11.07.2015 Views

The Structure of Complex Nominals: LIU, Hongyong A Thesis ...

The Structure of Complex Nominals: LIU, Hongyong A Thesis ...

The Structure of Complex Nominals: LIU, Hongyong A Thesis ...

SHOW MORE
SHOW LESS
  • No tags were found...

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

<strong>The</strong> <strong>Structure</strong> <strong>of</strong> <strong>Complex</strong> <strong>Nominals</strong>:Classifiers, Possessives and Relatives<strong>LIU</strong>, <strong>Hongyong</strong>A <strong>The</strong>sis Submitted in Partial Fulfillment<strong>of</strong> the Requirements for the Degree <strong>of</strong>Doctor <strong>of</strong> PhilosophyinLinguistics© <strong>The</strong> Chinese University <strong>of</strong> Hong KongNovember 2006<strong>The</strong> Chinese University <strong>of</strong> Hong Kong holds the copyright <strong>of</strong> this dissertation. Anyperson(s) intending to use a part or whole <strong>of</strong> the materials in the thesis in a proposedpublication must seek copyright release from the Dean <strong>of</strong> the Graduate School.i


e grammaticalized into Pr, but this grammaticalization path is not unitary. Besidesclassifiers, Pr may take on other forms such as possessive marker ge in Cantonese,possessive marker de in Mandarin, possessive marker a in Jingpo, and an empty possessivemarker in Nuosu Yi.When classifiers serve as RC markers in Nuosu Yi, we are able to tell the exactnumber <strong>of</strong> the relativized item. When a Nuosu Yi speaker does not want to specify theexact number <strong>of</strong> the relativized item, he will use the morpheme su. Semantically, it doesnot specify the number <strong>of</strong> the relativized noun, but it renders the relativized noun definite.We propose that this morpheme su is a definite article D, which plays a deciding role inencoding the definiteness <strong>of</strong> the whole complex nominal which contains a RC. <strong>The</strong> relativeclause marker, in this case, is a phonetically null element.<strong>The</strong> whole picture <strong>of</strong> the complex nominal structure in Yi is further complicated by thecooccurrence <strong>of</strong> su and a classifier with a RC. In this case, su is analyzed as a D-element.<strong>The</strong> real predication operator is still the classifier.<strong>The</strong> conclusion <strong>of</strong> this thesis is that possessive markers and relative markers areindeed <strong>of</strong> the same syntactic category Pr and they come into narrow syntax computation(from Numeration to LF) in the same manner.v


論 文 提 要本 文 討 論 的 是 存 在 於 複 雜 性 名 詞 性 短 語 結 構 中 的 主 謂 關 係 。 語 料 主 要 來 自 涼 山 彝語 。 基 於 涼 山 彝 語 做 出 的 分 析 , 同 樣 適 用 於 漢 語 普 通 話 、 廣 東 話 和 景 頗 語 。Bowers (1991: 19) 認 爲 漢 語 普 通 話 中 的 量 詞 從 語 義 功 能 的 角 度 來 講 是 將 性 質 轉化 爲 命 題 函 數 的 運 算 元 (Pr)。 基 於 他 的 分 析 , 我 們 認 爲 涼 山 彜 語 中 的 量 詞 也 可 以 看 成是 這 一 類 的 運 算 元 。量 詞 作 為 這 樣 的 運 算 元 存 在 於 涼 山 彜 語 的 領 屬 結 構 中 , 起 著 連 接 領 有 者 和 從 屬者 , 在 兩 者 間 建 立 領 屬 關 係 的 作 用 。 在 彜 語 的 領 屬 結 構 中 ,Pr 有 可 能 不 一 定 表 現 爲 量詞 。 當 不 表 達 從 屬 者 的 具 體 數 目 時 , 彝 語 便 不 使 用 量 詞 來 連 接 領 有 者 和 從 屬 者 。 在 這種 情 況 下 ,Pr 沒 有 語 音 形 式 。 基 於 這 樣 的 觀 察 , 我 們 認 爲 東 亞 量 詞 性 語 言 中 的 量詞 很 有 可 能 經 過 語 法 化 的 過 程 成 爲 Pr, 但 是 並 不 是 Pr 只 能 表 現 爲 量 詞 , 它 可 能 有 其他 的 表 現 形 式 。 例 如 在 廣 東 話 裏 它 可 以 表 現 為 領 屬 標 記 ge, 普 通 話 裏 為de, 景 頗語 裏 為 a, 涼 山 彝 語 裏 為 零 形 式 。當 彜 語 量 詞 作 爲 關 係 子 句 的 標 誌 時 , 它 能 顯 示 被 關 係 化 的 名 詞 的 單 複 數 特 徵 。 如果 被 關 係 化 的 名 詞 的 單 複 數 特 徵 不 清 楚 時 , 關 係 子 句 後 出 現 的 不 是 量 詞 , 而 是 另 外 一個 語 素 su, 它 能 使 得 被 關 係 化 的 名 詞 成 爲 一 個 定 指 名 詞 。 我 們 認 爲 su 是 彜 語 的 定 冠詞 。 它 在 確 定 包 含 關 係 子 句 的 複 雜 性 名 詞 結 構 的 有 定 性 方 面 發 揮 著 重 要 的 作 用 。 在 有su 的 關 係 子 句 結 構 中 , 關 係 子 句 的 標 記 和 領 屬 結 構 的 標 記 一 樣 , 是 零 形 式 。 涼 山 彜 語的 su 可 以 和 量 詞 同 時 出 現 在 一 個 名 詞 性 短 語 結 構 中 。 在 這 種 情 況 下 , 我 們 同 樣 認 爲su 是 定 冠 詞 , 而 量 詞 是 述 謂 運 算 元 (Pr)。在 考 察 了 漢 語 普 通 話 和 彜 語 的 領 屬 結 構 和 關 係 子 句 結 構 後 , 本 文 得 出 的 結 論 是 領屬 標 誌 和 關 係 子 句 的 標 誌 屬 於 同 一 句 法 範 疇 , 它 們 進 入 句 法 運 算 的 方 式 是 一 致 的 。vi


CHAPTER 7 CONCLUSIONS 227REFERENCES 239x


LIST OF ABBREVIATIONSAGENTAPARTASPCCHCLPCPDATDemPDOCDPEXPFPGENIPLCAMPNOMNumPOPPARPL/PlPOSSPossPPrPrPRCRC-MSCSFPSg/sgTBTOPTPAgent markerAdjective PhraseArticleAspect marker<strong>Complex</strong> Classifier HypothesisClassifier PhraseComplementizer PhraseDative CaseDemonstrative PhraseDouble Object ConstructionDeterminer PhraseExperiential aspect markerFunctional ProjectionGenitive markerInflection PhraseLinear Correspondence AxiomMinimalist ProgramNominative CaseCardinal Numeral PhraseOperatorParticlePlural markerPossessive morphemePossessive PhrasePredication operatorPredication PhraseRelative ClauseRelative Clause MarkerSmall ClauseSentence Final Particlesingular morphemeTibeto-BurmanTopic markerTense Phrasexi


CHAPTER 1Introduction1.1 Research objectives<strong>The</strong> main goal <strong>of</strong> this dissertation is to investigate the structure <strong>of</strong> complex nounphrases in four Sino-Tibetan classifier languages Mandarin, Cantonese, Jingpo and Yi,(two typical Tibeto-Burman (TB) languages), with a focus on the function <strong>of</strong> theclassifier in these languages. This dissertation, on the one hand, proposes a theory thataccounts for the predication relation holding within complex NPs in light <strong>of</strong> theparallelism existing between nominal structures (DP) and sentential structures (TP/CP) inlight <strong>of</strong> Abney (1987) and Szabolcsi (1994); on the other hand, this dissertation alsoproposes a theory that accounts for the encoding <strong>of</strong> specificity and definiteness incomplex noun phrases. Yi serves as a database for the formulation <strong>of</strong> the two theoreticalaccounts. What is proposed for Yi complex nominals will be extended to account forother Sino-Tibetan classifier languages for the verification <strong>of</strong> the proposed theories.1.2 A sketch <strong>of</strong> four Sino-Tibetan classifier languages<strong>The</strong> four Sino-Tibetan classifier languages to be discussed in this dissertation are Yi,Cantonese, Mandarin, and Jingpo.Yi, which was also called Lolo in the past, is mainly spoken in the southwesternprovinces <strong>of</strong> Sichuan, Yunnan, Guizhou and Guangxi in China. It is a TB language and ithas a large number <strong>of</strong> varieties across the three provinces. <strong>The</strong> total population <strong>of</strong> the Yinationality is around 6.57 millions. <strong>The</strong> total speaking population <strong>of</strong> the Yi languageshould be less than that, since some Yi people living in big cities with the Han prefer toteach and speak with their children Mandarin. <strong>The</strong> variety studied in this dissertation is1


Sichuan Liangshan Yi, which is a subdialect <strong>of</strong> northern dialect <strong>of</strong> Yi. 1 <strong>The</strong> data aremainly collected during my fieldwork in Liangshan Yi Autonomous Prefecture in SichuanProvince from October to December 2005. I have also drawn some examples fromreference grammar books. However, I have modified their transcriptions by adopting thestandardized Yi romanization scheme together with <strong>of</strong>ficial Yi orthography (ꆈꌠꁱꂷ)promoted in Liangshan.Cantonese belongs to the Yue group <strong>of</strong> Chinese dialects. It is spoken by about 100million people in the southern provinces <strong>of</strong> Guangdong and Guangxi and in neighboringareas such as Hong Kong and Macao, as well as throughout South-East Asia in Singapore,Malaysia, Thailand and Vietnam. <strong>The</strong> variety <strong>of</strong> Cantonese discussed in this dissertationis Hong Kong Cantonese. I will use Jyutping to present Cantonese data together withtraditional Chinese characters. 2Mandarin is a member <strong>of</strong> the Sino-Tibetan language family. <strong>The</strong> Chinese language(Hanyu), spoken in its Mandarin form, is the <strong>of</strong>ficial language <strong>of</strong> the mainland China andTaiwan, and one <strong>of</strong> four <strong>of</strong>ficial languages <strong>of</strong> Singapore. <strong>The</strong> government <strong>of</strong> China hasdeveloped a system <strong>of</strong> writing Chinese in the Roman alphabet, known as Hanyu Pinyin( 汉 语 拼 音 / 漢 語 拼 音 ‘Chinese spelling according to sounds’). <strong>The</strong> Mandarin used in thisdissertation refers to the variety spoken in the mainland. All Mandarin data are given inHanyu Pinyin together with simplified Chinese characters.Jingpo (also known as Kachin in Burma) is spoken by the Jingpo ethnic group who1 <strong>The</strong> examples in this dissertation are from the Shengzha dialect, which serves as a lingua franca inSichuan Liangshan Yi Autonomous Prefecture. It is the basis for the <strong>of</strong>ficial Sichuan Yi orthography (Table<strong>of</strong> Standard Yi Language Characters). <strong>The</strong> language is also called Nuosu, Nosu, Northern Yi, Sichuan Yi,or Liangshan Yi.2 Jyutping is the Cantonese Romanization Scheme developed by the Linguistic Society <strong>of</strong> Hong Kong.2


lives in China’s southwest province Yunnan along the Sino-Burmese border. 3<strong>The</strong>speaking population <strong>of</strong> Jingpo in Yunnan is about 100,000 and approaching 1.5 millionsacross the China-Burma border. <strong>The</strong> language has a written system modeled in Romanalphabets. Due to intensive language contact with Mandarin, Burmese, and other minoritylanguages in Yunnan, some new forms have emerged in Jingpo with some old formsbeing changed or replaced at a rapid pace. <strong>The</strong> language is known to be undergoing asignificant evolution from a non-configurational language (i.e. an agglutinating language)to an increasingly configurational language (i.e. an analytic language). Cheung (2003)has described Jingpo nominal structure in detail. Most <strong>of</strong> the Jingpo data used in thiswork are from Cheung (2003), Dai and Xu (1992), Gu (2005), Gu and Dai (2002), andgrammatical judgments from Jingpo native speakers.1.3 An overview <strong>of</strong> the dissertationIn Chapter 2, we argue that there exists a functional head Pr in complex nominalstructures. It is called Nm in Bowers (1991). <strong>The</strong> function <strong>of</strong> this head is to facilitatepredicativization by means <strong>of</strong> turning a saturated element (a property ) into anunsaturated element (a function ), and forming a predication between its specifierand complement. Like Bowers, we treat classifiers as the phonetic realization <strong>of</strong> thisfunctional head, but unlike Bowers, this PrP (NmP) is not the complement <strong>of</strong> D. Ratherthere is another functional projection FP between PrP and DP.We treat weak determiners in classifier languages as cardinality predicates ratherthan adjectival modifiers. We do not take possessor to be base generated at [Spec, DP],3 Note that the term ‘Jingpo’ is the <strong>of</strong>ficial Chinese Romanization <strong>of</strong> what is spelled elsewhere in theliterature as ‘Jinghpo’ or ‘Jinghpaw’.3


which is an A’-position. Instead, we adopt Szabolcsi’s idea that there is [N+I]P (PossP)between DP and PrP. Possessor reaches [Spec, PossP] via predicate inversion. Based onthese two assumptions and other empirical factors <strong>of</strong> Yi, we propose the <strong>Complex</strong>Classifier Hypothesis: a classifier forms a complex classifier with its complement: AP,DemP, or NumP. <strong>The</strong> resultant complex classifier (CL+AP/DemP/NumP) serves as thepredicate to predicate <strong>of</strong> the bare noun arguments at [Spec,CLP].(1)In Chapter 3, we observe that the distribution and interpretation <strong>of</strong> classifiers innominal structures may vary according to different syntactic properties <strong>of</strong> certainfunctional heads in different languages. <strong>The</strong> detailed difference in referential property <strong>of</strong>nominals can be seen through such strings as {classifier, N}, {Possessor, classifier,possessive head noun}, and {RC, classifier, relative head noun} in Mandarin, Cantoneseand Yi. With regard to Yi, there are two ways to encode definiteness. <strong>The</strong> first way,which is also employed by Mandarin but not by Cantonese, is through bare nouns. <strong>The</strong>second way is through the insertion <strong>of</strong> the definite article su.If a definite nominal contains CL, in general, two options are available to encodedefiniteness. <strong>The</strong> first one, represented by Cantonese, is to give all the deictic force to CL;thus Cheng and Sybesma’s (1999) analysis. <strong>The</strong> second option, exemplified by Yi andMandarin, is to let CL share the deictic force with other elements. In Mandarin, a4


demonstrative is required to encode definiteness; whereas in Yi the definite article sutogether with the classifier can encode definiteness. 4In Chapter 4, we examine possessives in Yi and Mandarin. <strong>The</strong> semantic relation <strong>of</strong>possession in Yi can be expressed in different nominal forms. Besides nominalpossessives, Yi also has lexical verbs to indicate possession. <strong>The</strong> most widely used verbis bbop (ꁨ), which is the counterpart <strong>of</strong> English possess or own. Some existential verbscan also be used to indicate possession. jjo (ꐥ) is a versatile verb and it can expresslocation, possession, and existence. We show that existence is a combination <strong>of</strong> locationand possession. We argue that an existential possessive sentence is derived from anexistential sentence with the possessor extraction.In Chapter 5, we examine the derivation and interpretation <strong>of</strong> relative clauses in Yiand Mandarin. In Yi, relative clauses can be marked either by the definite article su, or byclassifiers. We follow Kayne’s analysis <strong>of</strong> relatives. Instead <strong>of</strong> an IP that moves to[Spec,DP], as in English and Chinese, it is a CP that moves to [Spec,DP] in Yi. That is,all the material following D has to move to [Spec,DP]. We have also shown that su isoptional. When su is absent from a relative clause, a classifier will be incorporated into Dand perform the task <strong>of</strong> forming relatives, and mark the relative head as indefinite. <strong>The</strong>reis only one occasion where su can co-occur with a classifier: the relative head noun is4 <strong>The</strong> definite article in the Yi language su is different from the English definite article the in that su cannotgo directly with the noun. <strong>The</strong> Yi definite article su has to go together with a classifier.(i) the student (English)(ii) a. *ssoxsse su (Yi)ꌶ ꌺ ꌠstudent ARTIntended meaning ‘the student’b. ssoxsse ma suꌶ ꌺ ꂷ ꌠstudent CL ART‘the student’5


definite, and its number feature (plurality/singularity) is expected to be overtly specified.On that occasion, su and a classifier co-occur in the sentence and they jointly markrelative clauses.In Chapter 6, we examine Jingpo <strong>Complex</strong> <strong>Nominals</strong>. We focus on the distribution<strong>of</strong> Jingpo demonstratives. With respect to the distribution <strong>of</strong> Jingpo demonstrativeswithin the noun phrases, there is an asymmetry in distribution between the singular andplural demonstratives: i) the singular demonstrative can occur in either prenominal orpostnominal positions, while plural demonstratives are restricted to postnominal positions;ii) demonstratives, either singular or plural, can be inserted between the noun and theclassifier. Our analysis shows that this free positioning <strong>of</strong> demonstratives within nominalstructure is due to the dual status <strong>of</strong> demonstratives. When they occur in the prenominalposition, they are D. When they occur in the postnominal position, they are AP. Since Dhas no ability to encode plurality, plural demonstratives can never occur in the D. <strong>The</strong>ycan only serve as AP modifiers. <strong>The</strong>refore, they will invariably occur at the postnominalposition.1.4 <strong>The</strong>oretical AssumptionsThis thesis investigates the syntax <strong>of</strong> complex nominals within the framework <strong>of</strong> theMinimalist Program (Chomsky 1993, 1994). In this section I will spell out certain aspects<strong>of</strong> the theory directly relevant to this thesis. With regard to the referential property <strong>of</strong>nominals, the two key notions definiteness and specificity are also defined here as myworking assumptions.6


1.4.1 Definiteness1.4.1.1 DefinitesDefiniteness, as an important syntactic and semantic concept, has been viewed inmany different ways. One <strong>of</strong> them is known as Familiarity Hypothesis, which states thatthe English definite article the signals that the entity denoted by the noun phrase isfamiliar to both Speaker and Hearer; whereas the English indefinite article a/an does notsignal such shared familiarity. Lyons (1999) argues that “familiarity” is what enablesHearer to identify the referent. <strong>The</strong> referent <strong>of</strong> some definite noun phrase may beunfamiliar to Hearer, but as long as s/he is able to identify a referent for it, then the use <strong>of</strong>the definite noun phrase is successful. <strong>The</strong>refore Lyons replaces Familiarity byIdentifiability, and regards Identifiability as one <strong>of</strong> the two conditions for definiteness.Lyons has also observed that there are associative uses <strong>of</strong> definite NPs. For example,(2) I wonder who the anesthetist is.Although the hearer cannot identify the referent <strong>of</strong> the definite NP, since there is only oneunique anesthetist who will be involved in the operation, the use <strong>of</strong> definite NP issuccessful. <strong>The</strong> same logic can be extended to definite plural NPs and definite mass NPs:(3) I wonder who the anesthetists are.<strong>The</strong> hearer who hears this sentence cannot identify the referent <strong>of</strong> ‘the anesthetists.’However, the definite article is possible here, simply because the reference <strong>of</strong> the pluralNP is applied to the whole set <strong>of</strong> anesthetists. If some anesthetists referred to here are notincluded in the denotation <strong>of</strong> the set established here, then the definite plural NPs will fail.<strong>The</strong> following example gives a better illustration.(4) I’ve washed the dishes.7


If there are some dishes unwashed, that is, if the reference <strong>of</strong> ‘the dishes’ is not applied tothe whole set <strong>of</strong> 〖dishes〗, then (4) is false.Based on the above analysis, Lyons (1999) proposes a binary definition ondefiniteness.(5) a. Identifiabilityb. Inclusiveness (uniqueness for singular definite NPs)Heim and Kratzer (1998) defines the definite article as such that the determiner thedenotes a function with arguments in D and values in D e . For instance,(6) the opera by Beethoven〖the〗, which applies to the function 〖opera by Beethoven〗, yields Fidelio. <strong>The</strong>generalization is:(7) For any f ∈D such that there is exactly one x for which f(x)=1, 〖the〗(f) = theunique x for which f(x) = 1.This simply means that if there is exactly one x such that this x is a member <strong>of</strong> the setdefined by the function {x: x is an opera by Beethoven}, the denotation <strong>of</strong> the phrase “theopera by Beethoven” is the unique opera by Beethoven Fidelio. If Beethoven producedmore than one opera, or Beethoven did not produce any opera, then the phrase “the operaby Beethoven” denotes nothing at all. However, the concept <strong>of</strong> uniqueness is dependentupon contextual information. Take the following sentence as an example:(8) <strong>The</strong> dog is barking.In this sentence “the dog” does not indicate that there is only one dog in the whole actualworld, but it denotes the unique dog on the utterance occasion. <strong>The</strong>refore, Heim andKratzer (1998) revises the lexical entry for the English definite article the:8


(9) 〖the〗=λf: f ∈D and there is exactly one x ∈C such that f(x)=1.the unique y ∈C such that f(y) = 1,where C is a contextually salient subset <strong>of</strong> D.This means that the denotation <strong>of</strong> the is a function mapping a function f [defined as:<strong>The</strong>re is exactly one x, this x is a member <strong>of</strong> the set {x: x is a dog}.] to the unique y suchthat y is also a member <strong>of</strong> the set {x: x is a dog}. To put it simply, the denotation <strong>of</strong> thedefinite article used with a singular noun phrase presupposes the existence <strong>of</strong> a singleton<strong>of</strong> the type before mapping the denotation <strong>of</strong> the noun phrase (a function <strong>of</strong> type ) to a unique entity.If we assume the semantic type <strong>of</strong> the is < , e>, then we have to admit that thedomain <strong>of</strong> the function〖the〗is not D , but a subset <strong>of</strong> D . This means the mappingis just partial. <strong>The</strong> function is a partial, rather than total, function. <strong>The</strong> presuppositionresults from this partial function, since partial function assumes a one-membered subset(singleton) as the domain <strong>of</strong> the main function.This analysis can be extended to definite plural noun phrases as well as definitemass nouns, if we adopt the inclusiveness condition proposed in Hawkins (1978). 5 With adefinite noun phrase, the reference is to the only entity or the totality <strong>of</strong> the objects ormass in the form <strong>of</strong> a unique set. To sum up, there are two conditions a definite NP5 Chierchia (1998) uses the τ–operator to interpret the definite article, as indicated in the following:(a). τX=the largest member <strong>of</strong> X if there is one (else, undefined)(b). the dogs= τ DOGS=the largest plurality <strong>of</strong> dogs(c). the dog= τDOG=the only dog (if there is one)If the τ-operator applies to a set <strong>of</strong> pluralities (like the extension <strong>of</strong> the plural nouns dogs), it will refer tothe largest plurality in that extension (i.e., the one that comprises all <strong>of</strong> the dogs). If it applies to a set <strong>of</strong>singularities, it also seeks the largest one. <strong>The</strong> successful use <strong>of</strong> the singular definite article presupposesthat the predicate dog has only one object in its extension. Without this singularity presupposition, thedefinite singular article will yield an undefined result.9


should meet:(10) a. PRESUPPOSITION: a definite NP presupposes a set, which is a singleton. 6b. IDENTIFIABILITY: the denotation <strong>of</strong> this singleton should be identifiable.1.4.1.2 IndefinitesDiesing (1992) differentiates two types <strong>of</strong> indefinites: those that form restrictiveclause structures, and those that are bound by existential closure. <strong>The</strong> one forming arestrictive clause is called presuppositional (specific) indefinite; the one with existentialforce is called cardinal (existential, nonspecific) indefinite. For example:(11) a. Every llama ate a banana.b. Every x [x is a llama] ∃ y y is a banana ^ x ate yquantifier restrictive clausenuclear scope<strong>The</strong> universal quantifier does not quantify over everything, but over everything thatis a llama. <strong>The</strong> restriction on the quantifier is given in the restrictive clause [x is a llama].<strong>The</strong> quantifier every binds all the NP variables in the restrictive clause (the variable x inthis case). That is to say, the set defined by the restrictive clause can be taken to representthe existence presupposition induced by the quantifier. Existential closure in turn bindsall the remaining variables in the nuclear scope (the variable y in this case).Based on this tripartite structure <strong>of</strong> logical form, Diesing proposes MappingHypothesis:6 For definite plural NPs and definite mass NPs, the member <strong>of</strong> the set denoted by them may not be one, butwe can still regard it as a singleton in an extensive meaning, since the reference <strong>of</strong> definite plural NPs anddefinite mass NPs is to the totality <strong>of</strong> the set.10


(12) Mapping Hypothesisa) Material from VP is mapped into the nuclear scope.b) Material from IP is mapped into a restrictive clause.Put it simply, [Spec, VP] is the position for nonspecific indefinites, because it is inthe nuclear scope, and it can only be bound by the existential closure; whereas [Spec, IP]is the position for specific indefinites, because it forms a restrictive clause bypresupposing a reference set.<strong>The</strong> existential type <strong>of</strong> indefinites lacks quantificational force <strong>of</strong> their own, 7 so theycan only get existential force from existential closure. <strong>The</strong> presuppositional type <strong>of</strong>indefinites has quantificational force and form operator-variable structures by introducinga restrictive clause. To sum up, indefinites include two types according to the prenominaldeterminers:(13) a. PRESUPPOSITIONAL: form operator-variable structuresb. CARDINAL: form cardinal predicatesIn English, indefinites are actually ambiguous between presuppositional andnon-presuppositional (cardinal) readings.Strong determiners and weak determinersMilsark (1974) distinguishes two types <strong>of</strong> determiners, which he calls strong andweak determiners. <strong>The</strong> best-known diagnostic to differentiate them is that weakdeterminers can appear in there construction (Milsark 1974, Bowers 1991, Diesing 1992,7 <strong>The</strong> pre-nominal weak determiners (cardinal numerals) can only serve as cardinal predicate; therefore,they play no role in defining the quantificational property.11


Heim and Kratzer 1998). 8 In such a case, only cardinal reading is possible.(14) <strong>The</strong>re are some/many/a few/three flies in my soup.(15) <strong>The</strong>re is a fly in my soup.(16) * <strong>The</strong>re are all/most flies in my soup.(17) * <strong>The</strong>re is the/every fly in my soup.<strong>The</strong> distinction between these two types <strong>of</strong> determiners results from theirpresuppositional property. Strong determiners presuppose the existence <strong>of</strong> the entitiesthey are applied to; whereas weak determiners are ambiguous between a presuppositionalreading and a non-presuppositional reading. By non-presuppositional reading, we meanthat they merely assert, rather than presuppose, the existence <strong>of</strong> whatever entities they areapplied to.According to Heim and Kratzer (1998), presupposition means partial functionsrather than total functions. Partial functions refer to the mapping from a subset to anotherset rather than a set-to-set mapping. <strong>The</strong> existence <strong>of</strong> a subset reflects the existence <strong>of</strong> apresupposition. For example,(18) <strong>The</strong>re are some ghosts in my house.(Existence assertion: some has the cardinal reading.)(19) Some ghosts are in the pantry; the others are in the attic.(Existence presupposition: some has the presuppositional reading.)Presuppositional NPs are raised higher up by Quantifier Raising Rule, and formrestrictive clauses, and have the whole IP as their scope domain; therefore they can havewider scope than cardinal NPs. Cardinal NPs can only receive existential force byexistential closure. According to Bowers (1991), strong determiners belong to the8 <strong>The</strong> following examples are taken from Diesing (1992: 59) with minor modifications.12


category D, whereas weak determiners are simply adjectives. <strong>The</strong> following are sometypical strong ad weak determiners.(20) a. Strong determiners: each, every, all, most…b. Weak determiners: many, several, some, a few…1.4.2 SpecificityVarious assumptions have been made about specificity. <strong>The</strong> most widespread view isthe one that an NP is specific if it has wide scope over an operator (including quantifierNPs, negation operators, modals, and propositional attitude verbs). However, empiricalevidence shows that specific NPs may allow narrowest scope. <strong>The</strong>refore, Enç (1991)argues that specificity must be defined independent <strong>of</strong> scope relations. In Turkish,specificity is marked morphologically by the accusative case marker –(y)i (data takenfrom Enç: 1991):(21) a. Ali bir kitab-i aldi. (specific indefinite)Ali a book-acc bought‘A book is such that Ali bought it.’b. Ali bir kitap aldi. (non-specific indefinite)Ali a book bought‘Ali bought some book or other.’In (21a) the object is marked with accusative case, and it can only have a specificreading; in (21b) there is no case marker on the object NP, and it has a nonspecific13


(existential) reading. Unlike English, Turkish indefinite objects are not ambiguousbetween specific and nonspecific readings.Following Diesing’s (1992) Mapping Hypothesis, the essential semanticcontribution <strong>of</strong> “specificity” is presuppositionality. Strong determiners always assume theexistence <strong>of</strong> presuppositions. If the presupposed subset is a singleton, then it isdefiniteness. If the number <strong>of</strong> the presupposed subset is more than one, then it is specificindefiniteness. If there is no presupposition, then it is nonspecific indefiniteness. Hencemy working definition <strong>of</strong> definiteness and specificity can be best illustrated as follows:(22)14


If there is no presupposition for an NP, this NP is nonspecific indefinite. If there ispresupposition for an NP, then it is specific. If the number <strong>of</strong> the members <strong>of</strong> thepresupposed set is just one, then it is definite. If the number is more than one, then it isspecific indefinite. Here one important notion is in need. To illustrate, consider (23):(23) I read the three books on the table.[ DP the three books] presupposes a set with three members, and according to ourdefinition, it should be a specific indefinite NP, but it is a definite NP, simply because ‘thethree books’ are used as an inseparable unit, and in this sense ‘the three members’ form aset and this single set is the only member <strong>of</strong> the presupposed set. This is exactly what“inclusiveness” means. <strong>The</strong>refore, we cannot say (24), although specific indefinite in (25)is perfectly acceptable.(24) * I read the three books on the table, and there is one book on the same table leftunread.(25) I read three books on the table, and there is one book on the same table left unread.Nonspecificity and Existentiality<strong>The</strong> difference between specific indefinites and nonspecific indefinites lies in thefact that specific indefinites presuppose existence, whereas nonspecific indefinites assertexistence. <strong>The</strong>refore, specifics that presuppose existence are blocked in existentialsentences, since existential sentences assert existence rather than presuppose existence.Nonspecificity and cardinal predicatesHigginbotham (1987) addresses the problem <strong>of</strong> predicate nominals in English.15


(26) John is a lawyer.If “a lawyer” is regarded as a DP, then it is an entity, which is a saturated concept.<strong>The</strong>refore, it cannot serve as a predicate, since a saturated element has no place to befilled in. However, (26) is grammatical. Based on this observation, Higginbothamproposes the following representation for the indefinite NP “a lawyer”(27) a (x) & lawyer (x)According to Higginbotham, modification is interpreted as conjunction. For example,“brown cow” will apply truly to the things that are brown and cows.(28) brown (x) & cow (x)Higginbotham interprets the indefinite article a simply as an adjective. <strong>The</strong> same logiccan be extended to numerals, that is, cardinal numerals are adjectives as well.(29) <strong>The</strong>y are three friends <strong>of</strong> mine.<strong>The</strong> predicate is interpreted as in (30),(30) three (x) & friends <strong>of</strong> mine (x)This complex predicate is used to predicate <strong>of</strong> the subject “they”. Here, numerals are usedas cardinal predicates.(31) <strong>The</strong> cars were three (in number). 9We would like to argue that although we can treat cardinal numerals as adjectives, theyare different from true adjectives. Higginbotham’s interpretation (30) is, in fact, aninaccurate formula. <strong>The</strong> cardinal number is not predicated <strong>of</strong> the variable x; rather it ispredicated <strong>of</strong> the number <strong>of</strong> x. <strong>The</strong>refore, we modify it as follows:(32) a. three (x) & friends <strong>of</strong> mine (x)b. three (/x/) & friends <strong>of</strong> mine (x)9 Example (31) is from Bowers (1991: 18).16


Like Higginbotham, we assume that cardinals are predicates, but unlike him, we donot think that cardinal numerals stand in conjunction with a noun phrase, since the formerpredicates <strong>of</strong> a hidden subject the number <strong>of</strong> my friends, and the latter predicates <strong>of</strong> thereal subject they.1.4.3 MP and Feature Checking <strong>The</strong>ory1.4.3.1 Guiding Ideas <strong>of</strong> the MPTwo major tasks <strong>of</strong> generative grammar are to find a way to account for thephenomena <strong>of</strong> particular languages (descriptive adequacy) and to explain how knowledge<strong>of</strong> these facts arise in the mind <strong>of</strong> the speaker-hearer (explanatory adequacy) (Chomsky1994). In terms <strong>of</strong> language acquisition, the two tasks can be rephrased as characterizinglanguages (states) attained (descriptive adequacy) and the shared initial state (explanatoryadequacy) (Chomsky 2001).Since language is a recursive system, the computation must contain at least Merge,which takes two syntactic objects and form a new object. <strong>The</strong> guiding principle formerging is referred to as Agree in MP. “<strong>The</strong> empirical facts make it clear that there are(LF) uninterpretable inflectional features that enter into agreement relations withinterpretable inflectional features…<strong>The</strong> obvious conclusion, which we adopt, is that theagreement relation removes the uninterpretable features from the narrow syntax, allowingderivations to converge at LF while remaining intact for the phonological component(with language-variant PF manifestation)…we therefore have a relation Agree holdingbetween a and b, where a has interpretable inflectional features and b has uninterpretableones, which delete under Agree” (Chomsky 1999: 3). In this sense, the minimalist17


program (MP) can be viewed as a feature-checking program.Besides Merge, other operations are needed to derive all the sentences in a language.Among these operations is Move, which deals with the prevalent ‘displacement’phenomena in any particular language. Displacement is implemented by selecting a targetP and a related category K to be moved to a position determined by P, which is a probethat seeks K. <strong>The</strong> probe P determines the kind <strong>of</strong> category that can be moved to thisposition.1.4.3.2 Locality conditionGiven the operation <strong>of</strong> Move, it is clear that one <strong>of</strong> the most important questions <strong>of</strong>syntax is to work out what economic conditions are functioning to constrain thederivation from the lexicon to PF and LF. This thesis will assume the Local Economy<strong>The</strong>ory proposed by Collins (1997), which states that there are only two real economyconditions, both <strong>of</strong> which are local:(33) a. Last Resort: an operation involving a may apply only if some property <strong>of</strong> a issatisfied.b. Minimality: an operation (satisfying Last Resort) may apply only if there is nosmaller operation (satisfying Last Resort).It can be restated as a requirement that operations satisfying Last Resort should beminimal.18


1.4.3.3 LCA and Bare Phrase <strong>Structure</strong>1.4.3.3.1 Kayne's Antisymmetry Syntax (1994)Kayne (1994) abandons the standard assumption that hierarchical structure andlinear order are independent <strong>of</strong> one another. He proposes to derive the linear order <strong>of</strong>terminal symbols from the hierarchical relations between the nonterminal symbolsdominating them. He starts out by noting that the linear ordering relation <strong>of</strong> a given set <strong>of</strong>terminal elements has three properties. It is transitive, i.e., if X > Y and Y > Z then X > Z;it is total, i.e., it covers all members <strong>of</strong> a given set <strong>of</strong> elements; and it is antisymmetric,i.e., if X > Y then it is not possible that Y > X. Based on these three properties, Kayneproposed Linear Correspondence Axiom (LCA): d(A) is a linear ordering <strong>of</strong> T.A is a set <strong>of</strong> ordered pairs <strong>of</strong> nonterminals such that the first nonterminalasymmetrically c-command the second. e.g. A={, , , }T is a set <strong>of</strong> terminals.d is a many-to-many mapping relation from non-terminals to terminals.d(A) is a set <strong>of</strong> ordered pairs <strong>of</strong> terminals mapped from the correspondingordered pairs <strong>of</strong> nonterminals.According to this axiom, the following tree (in accordance with X-bar theory) willbe out:(34)19


M asymmetrically c-commands R, which means that the pair is in the set <strong>of</strong>linear ordering relations. But here P also asymmetrically c-commands Q, which meansthat the pair is also in the set <strong>of</strong> linear ordering relations. This is contradictory: it isnot possible for and to be in the set <strong>of</strong> linear ordering relations <strong>of</strong> a given tree.If the tree is modified to the following, this problem disappears.(35)Adopting the segment/category distinction proposed by May (1985), Kayne definesc-command as follows: X c-commands Y iff X and Y are categories and X excludes Y (nosegment <strong>of</strong> X dominates Y) and every category that dominates X dominates Y (Kayne1994: 16). <strong>The</strong> top node L in (34) is replaced with P in (35). P is now a categoryconsisting <strong>of</strong> two segments. In (35) M still asymmetrically c-commands R, because everycategory that dominates M (i.e. P) dominates R. However, P no longer asymmetricallyc-commands Q, because P does not exclude Q, since the segment <strong>of</strong> the higher Pdominates Q. <strong>The</strong> pair is therefore no longer part <strong>of</strong> the set <strong>of</strong> linear orderingrelations, and the contradiction is eliminated.As Kayne points out, this discussion leads to the conclusion that specifiers andadjuncts cannot be distinguished. (34) shows that under the LCA a structure in which anon-head is the sister <strong>of</strong> another non-head cannot exist because such a configurationcannot be linearized. <strong>The</strong> only way in which two non-heads can be sisters is in theconfiguration where one is adjoined to the other. This means that specifiers must be20


adjoined; in other words, specifiers are adjuncts. LCA also tells us that every maximalprojection can allow only one specifier. Multiple specifiers are disallowed. We canillustrate this point with the phrase marker shown in (36).(36)LPPK M Pk Q R Sq r TtIn (36) the nonhead M has adjoined to the nonhead P, and the nonhead L has alsoadjoined to P. From the tree we can observe that L asymmetrically c-commands Q, and atthe same time M asymmetrically c-commands K. <strong>The</strong>refore, both and arein d(A), which violates antisymmetry. Since the head <strong>of</strong> a phrase marker plays amediating role in comparing the relative word order <strong>of</strong> the specifier and complement,Kayne proves that specifier and complement are always on opposite sides <strong>of</strong> the head.Otherwise, the antisymmetry cannot be established. <strong>The</strong> reason that we can have the wordorder <strong>of</strong> Specifier-Complement-Head (e.g., <strong>of</strong> SOV) in natural languages is that thecomplement has raised up to some specifier position to the left <strong>of</strong> the head. <strong>The</strong> typicalphrase marker that satisfies LCA is (37):(37)21


1.4.3.3.2 Chomsky’s Bare <strong>The</strong>ory (1994)Chomsky (1994) evaluates Kayne’s proposal and argues that Kayne’s proposal isvery much in the spirit <strong>of</strong> the minimalist program and can be incorporated into the barephrase structure theory. Chomsky’s bare theory holds that a lexical item can be both anX 0 and an XP. Chomsky takes clitics as an illustration. In its theta-position, a clitic is anXP; attachment to a head requires that it be an X 0 .(38)<strong>The</strong>se two trees given by Kayne can be interpreted as follows on the bare phrasestructure assumption:(39)Both Kayne’s admissible and inadmissible trees in (38) are reduced to (39). In thiscase no ACC holds between {m, p}. No ordering is assigned to m and p, and the structurefails LCA. However, Chomsky argues that LCA can ignore traces which will disappear atPF. If the terminal p in (39) is a trace, then (39) is legitimate. 1010 That is, if the complement is a single-terminal XP, then it must raise overtly. If XP = DP, then its head Dis a clitic, which attaches at a higher point. If XP = NP, then N must incorporates to V (Chomsky 1994: 28).22


1.5 SummaryIn this chapter, we introduce the research objectives and languages to be discussedin this dissertation. An overview <strong>of</strong> this dissertation is given chapter by chapter. We havealso presented our definition <strong>of</strong> definiteness and specificity as our working assumptions.Finally we give a sketchy introduction to the LCA and the minimalist framework as ourtheoretical framework for the whole dissertation.23


CHAPTER 2Predication and <strong>Complex</strong> Classifier Hypothesis2.1 PredicationIn the theory developed here, it is argued that there exists a functional head Pr innominal structure, along the lines suggested in Bowers (1991, 2001). Predication isthe process by which a predicate is combined with a subject in order to form aproposition. Predication is a relation holding between a predicate and its arguments.<strong>The</strong> process <strong>of</strong> combining a predicate with its arguments is called saturation. At thesentential level, saturation results in a clause with a subject and a predicate. If apredicate occurs in a complex nominal structure, its argument has to occur within thesame nominal structure. In this case we can get the nominal-internal predication,which is the major research target for this chapter.2.1.1 SaturationA complete sense always involves a logical combination <strong>of</strong> parts into a whole,which is a matter <strong>of</strong> saturating something unsaturated. <strong>The</strong>re are two things that aresaturated: truth-values and referents <strong>of</strong> DPs. <strong>The</strong>y are self-contained primitivesTruth-values are the extensions <strong>of</strong> propositions, also known as denotations orreferents <strong>of</strong> sentences. Saturation, as a process, is the application <strong>of</strong> a function to itsarguments. This saturation process is exactly the act <strong>of</strong> predication. Saturateddenotations include entities (e) and truth values (t). Saturated elements are used asarguments, and unsaturated elements are predicates.24


2.1.2 Bowers’ Predication <strong>The</strong>oryBowers (1991, 1993, and 2001) argues that the main clause (MC) and smallclause (SC) predication relation can be uniformly represented as follows:(1) PrPDP (subject)Pr’PrXP (predicate)X={V, A, N, P} (Bowers 1991)For MC, PrP is the complement <strong>of</strong> I; while for SC, it is the complement <strong>of</strong> V,leading to a complex verb phrase. Bowers (2002) explains that the functional categoryPr is a generalization <strong>of</strong> the “light verb” v, which has two inherent properties: (i) itsspecifier hosts the external argument <strong>of</strong> transitives and unergatives; (ii) itscomplement contains a probe with object agreement features, and that there is acategory Tr, encoding transitivity relation, between Pr and V. It is said that Tr containsphi-features and assigns accusative Case, as well as containing an EPP-feature. [Spec,Tr] is the position to which accusative Case-marked NPs are raised. <strong>The</strong>refore, Trreflects the transitivity <strong>of</strong> the corresponding predicate.25


(2)Because Tr has the strong EPP-feature, direct objects, which are widely assumedto be base-generated at [Spec,VP] (Larson 1988; Bowers 1991, 1993; Hale andKeyser 1993, and Collins 1997),will move to [Spec,TrP]. 1Semantically, Bowers (1991) follows Chierchia’s (1985) semantic theory, whichincludes four basic types: u, p, π, e plus the predication relation ∪ :π→, and itsreverse, the nominalization relation ∩ : →π. p is the type <strong>of</strong> proposition; π isthe type <strong>of</strong> property; and u is the type <strong>of</strong> basic entities. Predication in this theoryrefers to the operation <strong>of</strong> turning a property into a propositional function, which mustcombine with an entity to form a proposition. 2 <strong>The</strong> function <strong>of</strong> Pr is to map propertiesinto propositions.1 In the structure proposed in Bowers (1993, 2002), there are only head-to-head and spec-to-specmovements.2 <strong>The</strong>re is a clear difference between the three concepts predication, saturation, and predicativization.Predication refers to the process <strong>of</strong> turning a property into a function. Saturation refers to theapplication <strong>of</strong> this function to its arguments and the establishment <strong>of</strong> a predication relation.Predicaticativization refers to the creation <strong>of</strong> a subject position normally by means <strong>of</strong> either typeshifting or by operator movement. Viewed in this way, predication can be regarded as a sub-type <strong>of</strong>predicativization.26


Bowers (1991, 1993) argues that propositions will universally have a uniformsyntactic structure (and type assignments) <strong>of</strong> the form shown in (3):(3)In this phrase marker the direct object is regarded as the secondary subject, thusasymmetrically c-commands the complement <strong>of</strong> the verb. Regarding direct objects assecondary subjects captures an interesting similarity existing between PrP and VP,both involving applying an entity-denoting NP/DP (with the semantic type ) to afunction (propositional function for PrP and property function for VP) to form asaturated concept (proposition for PrP and property for VP). What is especiallyimportant is that direct objects are base-generated in [Spec, VP], parallel to theposition <strong>of</strong> subject in [Spec, PrP]. V-to-Pr movement is obligatory. <strong>The</strong> motivation forthe movement has two accounts. <strong>The</strong> first is to assume that Pr has a strong feature[+predicative] on the observation that only predicative expressions can be thecomplement <strong>of</strong> Pr. <strong>The</strong> second is based on the theta-role assignment. This motivationis discussed in detail in Japanese light verb structure in Grimshaw and Mester (1988).Light verb can inherit the theta-marking ability from the main V after V-to-v27


movement. That means that the lexical verb transfers its agent theta-marking ability tothe light verb. <strong>The</strong>refore, light verbs can assign agent role to the external argument. Inthe similar vein, Pr inherits the theta-marking ability from the lexical V after V-to-Prmovement. Either account suffices to explain the motivation for the V-to-Prmovement.2.1.3 DP-internal predicationIn light <strong>of</strong> the sentential predication theory, Bowers (1991) argues for theDP-internal predication and proposes the syntactic characterization <strong>of</strong> nominalstructures:(4)<strong>The</strong> category D has been argued for on syntactic grounds by Abney (1987).Bowers (1991) argues for the existence <strong>of</strong> a grammatical category Nm, intermediatingbetween D and N, which is parallel in form and content to the grammatical categoryPr.28


Like VP, AP, and PP, NP in this universal nominal structure denotes properties,and Nm is the predication operator, substantiating the predication relation by turning aproperty into a predicate . This DP-internal predication is in strict conformitywith the syntax <strong>of</strong> clausal predication discussed in 2.1.2.This universal nominal structure enables Bowers (1991) to relate the gerundivesubject in (5a) to the structure illustrated in (5b). 3(5) a. [John’s singing songs] is terrifying everyone.b.In this structure, [Spec, PrP] is not a case-marked position. Hence, an NP in thatposition must move to a position where it can receive Case. <strong>The</strong> only possibleposition is [Spec, DP], where it will be assigned genitive Case, thereby satisfying theCase Filter. This movement is the exact counterpart <strong>of</strong> the movement from [Spec, PrP]to [Spec,IP].3 Bowers (1991) argues that possessive NPs are either base-generated in [Spec, DP] or moved from[Spec, NmP], the canonical position for ‘subjects’ <strong>of</strong> nominals, or from [Spec, NP], the canonicalposition for ‘objects’ <strong>of</strong> nominals.29


Similarly, Bowers argues that action nominals derive from underlying structureswith real subject or object arguments in [Spec, NmP] and [Spec, NP] and the headNoun raises to Nm, parallel to the raising <strong>of</strong> V to Pr in clauses. For action nominals,the theme argument can occur as genitives.(6) the destruction <strong>of</strong> the city/the city’s destruction(7) the publication <strong>of</strong> the article in the Times/the article’s publication in the Times(8)Result nominals and basic nouns, in contrast, neither have real subject and objectarguments, nor do they get raised to Nm. <strong>The</strong>y are arguments. For result nominals, thetheme argument cannot occur as genitives.(9) the student <strong>of</strong> Chemistry/*Chemistry’s student(10) the pro<strong>of</strong> <strong>of</strong> the theorem in the journal/*the theorem’s pro<strong>of</strong> in the journal30


(11)<strong>The</strong> difference between action nominals and result nominals lies in the fact thatthe object <strong>of</strong> an action nominal is a real secondary subject, base generated at [Spec,NP], while the object <strong>of</strong> a result nominal is a SC-complement <strong>of</strong> N. That is, actionnominals assign theta-roles in exactly the same way as verbs do, while basic nounsand result nominals simply do not assign theta-roles at all. <strong>The</strong>refore, action nominalswould have to get raised to Nm for exactly the same reason that verbs obligatorilyraise to Pr, namely, to assign a theta-role to the primary subject in [Spec, NmP].Basic nouns and result nominals would not move because they do not have anytheta-roles to assign. <strong>The</strong> reason to assume the projection <strong>of</strong> NmP for result nominalsis to accommodate other prenominal adjectives, which function as the predicate <strong>of</strong> theresult nominal or the common noun.31


Now we can see that the nominal structure (1) is an extension <strong>of</strong> the semanticrelation defined at the sentential level by Bowers (1991), repeated below with someminor changes for ease <strong>of</strong> exposition.(12)Bowers (1991) differentiates two types <strong>of</strong> quantifiers in English:(13) D-quantifiers (strong determiners): all every, each, both, some, neither, anyA-quantifiers (weak determiners): many, few, several, much, two<strong>The</strong> difference is reflected in their syntactic and semantic function. Strongdeterminers (determiner quantifiers) map properties onto sets <strong>of</strong> properties, therebychanging the type <strong>of</strong> NmP: for example: all men32


Although weak determiners are regarded as adjectival phrases modifying NmP,they are different from other types <strong>of</strong> attributive APs. Bowers instantiated this pointby using data from Chinese. In Chinese, weak determiners precede classifiers, whileother types <strong>of</strong> attributive adjective phrases will be marked either by the so-calledmodification marker –de or marked by zero morpheme. For example:(16) san ben youqu de shu三 本 有 趣 的 书three CL interesting DE book‘three interesting books’(17) san ge yuan panzi三 个 圆 盘 子three CL round plate‘three round plates’In (16) the cardinal numeral san ‘three’ must be followed by the classifier ben, whilethe real adjectival modifier youqu ‘interesting’ must be followed by -de. In (17) thecardinal numeral san ‘three’ must be followed by the classifier ge, while the realadjectival modifier yuan ‘round’ is marked by zero morpheme.Based on this observation, Bowers proposes that classifiers are phoneticrealization <strong>of</strong> Nm.34


(18)Zhangsan-de na san ben youqude shuZhangsan-GEN that three CL interesting book‘those three interesting books <strong>of</strong> Zhangsan’s’Following the predication theory proposed by Bowers (1991), the difference betweenclassifier languages such as Mandarin and Yi and article languages such as English isthat in the latter Nm is a predication operator with a zero morpheme, whereas in theformer Nm is phonetically realized as classifiers.(19)35


<strong>The</strong> semantic function <strong>of</strong> this functional category Nm (nominal Pr) is to turnproperties into propositional functions. This work assumes the framework outlinedabove, and agrees with Bowers that Chinese classifiers are the phonetic realization <strong>of</strong>the functional head Nm, and the idea that classifiers could be the phonetic realization<strong>of</strong> nominal Pr will also be extended to other Sino-Tibetan classifier languages in thefollowing chapters. However, this work departs from Bowers in two significant ways.First, we do not treat weak determiners as modifiers <strong>of</strong> Nm. Instead we treat weakdeterminers as the complement <strong>of</strong> Nm. Second, we treat adjectival modifiers in thesame way as weak determiners. This means if numerals are used as cardinal predicate,they are not to be treated as adjuncts. Rather they occupy the complement position <strong>of</strong>NmP, serving as predicates to predicate <strong>of</strong> the subject at [Spec, NumP]. In thefollowing section, we are going to illustrate the problems with Bowers’ analysis <strong>of</strong>Chinese nominal structure.2.1.4 Bare nominals as arguments<strong>The</strong> reason why we treat weak determiners and APs as complement <strong>of</strong> CL ratherthan adjunct <strong>of</strong> CL lies in the unique syntactic and semantic properties <strong>of</strong> Chinesebare nominals. Bowers treats Chinese bare nominals in the same way <strong>of</strong> English barenominals, thus failed to observe the difference existing between English and Chinese.Cross-linguistically, bare noun phrases appear to have a double nature. In theirrole as quantifier restrictor (as in every man) and in predicate position (as in John andBill are doctors) they seem to be predicates. In their role as devices for kind reference36


(names <strong>of</strong> kinds), they must be arguments.Chierchia (1998) examines the semantic property <strong>of</strong> bare nominals from across-linguistic point <strong>of</strong> view and sets up a typology <strong>of</strong> NP denotations. He proposesthat in some languages (like Chinese), NPs are argumental (names <strong>of</strong> kinds) and canthus occur freely without determiner in argument position; in others they arepredicates (Romance), and this prevents NPs from occurring as arguments, unless thecategory D(eterminer) is projected. Finally, there are languages (like Germanic orSlavic) which allow both predicative and argumental NPs; these languages, being the‘union’ <strong>of</strong> the previous two types, are expected to behave like Romance for certainaspects <strong>of</strong> their nominal system (the singular count portion) and like Chinese forothers (the mass and plural portions).In the first type languages ([+arg, -pred]) such as Chinese, every N and itsphrasal projection NP is <strong>of</strong> type . All nouns start out as kind denoting. Bare nounswill be allowed to occur freely as arguments. This naturally leads to a problem. As weknow, determiners, being quantificational devices, want a restriction (i.e., a predicate).Since determiners look for predicates, but NPs denote kinds in such languages, weseem to have a type mismatch problem. Chierchia solves the problem by assumingthat in argumental languages, determiners will have to apply to kinds, which arereadily turned into predicates by a covert predication operator, since it is a fact that toany property there corresponds a kind, and every kind will have a correspondingproperty (the property <strong>of</strong> being an instance <strong>of</strong> that kind). <strong>The</strong> correspondence betweenproperties and kinds suggests that there are ways to get one from the other:37


predication operator and nominalization operator.Since the property is derived from the kind, the property can only have massdenotation. 4 <strong>The</strong>refore, no plural marking is allowed, and numerals will not be able tocombine directly with nouns: a classifier is obligatory.In the second type <strong>of</strong> languages ([-arg, +pred]) such as French, every N and itsphrasal projection NP is a predicate <strong>of</strong> type . Such languages disallow barenominal arguments (be it bare plurals or mass) altogether. In such a language, NPcannot be made into an argument without projecting D. D is obligatory.In the third type <strong>of</strong> languages ([+arg, +pred]) such as English, NPs can freely beeither predicative or argumental. Lexical nouns can either denote kinds or predicates,and their types can be freely shifted back and forth by the available type shiftingoperators. In English, N <strong>of</strong> the feature [+arg] will yield mass nouns; N <strong>of</strong> the feature[+pred] will yield count nouns. Mass nouns can occur as bare arguments. <strong>The</strong>nominalization operator will apply to plurals and generate a kind, which can be usedas bare arguments.If Chierchia’s typology <strong>of</strong> NP denotations is on the right track, then Bowers’treatment <strong>of</strong> Chinese nominal structure cannot hold, since Chinese NP has thesemantic type <strong>of</strong> , whereas English NP has the semantic type <strong>of</strong> . 5Return back to Bowers’ proposal in (18), we find it inappropriate to let Prsubcategorize for NP, since NP is not a property in Chinese. Rather, NP is argumental,4 Mass nouns are the neutralization <strong>of</strong> the singular/plural distinction. Chierchia (1998) argues that aproperty corresponding to a kind comes out as being mass, because the property <strong>of</strong> being an instance <strong>of</strong>a kind does not differentiate between singular and plural instances. Fido is as good an instance <strong>of</strong> thedog-kind as Fido and Barky are.5 Here we restrict English NPs (<strong>of</strong> the semantic type ) to the noun phrases whose heads are countnouns. For English mass noun, the semantic type is .38


Here the predicate three represents a property, indicating the ‘three-ness’ <strong>of</strong> thenumber <strong>of</strong> shu (‘book’). Nm turns the property <strong>of</strong> ‘three-ness’ into a function .This function is applied to the argument shu (‘book’), yielding a truth value p. Sincethere is no strong determiner in D, the default value <strong>of</strong> D will give rise to theexistential reading <strong>of</strong> the whole nominal. In this case the empty D is automaticallyinterpreted as the existential quantifier. <strong>The</strong> surface word order <strong>of</strong> san ben shu, wewill show later, is derived from (20) by predicate inversion movement.2.1.6 Additional evidence for treating CL as PrAccording to Bowers (1991), for MC, PrP is a complement <strong>of</strong> I; while for SC, itis a complement <strong>of</strong> V. We have shown that CL can be used as Nm in 2.1.5. Can CL beused as Pr at sentential level? It is not easy to find such data, since classifiers do notnormally enter into sentential level. However, we find that the typical genericclassifier <strong>of</strong> Mandarin ge does have such a function.(21) a. ta he le ge mingding da zui他 喝 了 个 酩 酊 大 醉 。he drink ASP CL completely drunk'He was drunk completely.'40


. ta ba zhe ping jiu he le ge ganjing.他 把 这 瓶 酒 喝 了 个 干 净 。He BA this CL wine drink ASP GE clean'He drank this bottle <strong>of</strong> wine up.'We can analyze the sentences into the following:c. ta i he le [pro i ge mingding da zui].d. ta ba [zhe ping jiu] i he le [pro i ge ganjing].In (21c) and (21d), the verb he (‘drink’) takes as its complement a small clause (PrP),which is headed by the CL ge. However, it should be noted that only the genericclassifier ge is allowed in such construction. 62.2 <strong>Complex</strong> Classifier HypothesisFollowing Chierchia (1998), this work assumes that nouns in Sino-Tibetanclassifier languages come out <strong>of</strong> the lexicon as fully referential arguments, since theseclassifier languages are [+arg, -pred] languages. Bare nouns are names <strong>of</strong> kinds. If weneed to use nouns in different ways rather than kind-denoting, we have to useclassifiers. I will refer to bare nouns as simple nominals, and noun phrases with otherelements complex nominals.Within complex nominals, Dem(onstrative)P, Num(eral)P, or A(djective)P firstcombines with CL to form a function, which is applied to the NP at [Spec, CLP],6 It remains a question for further research why only the generic classifier ge can be used in this way.Other classifiers are by no means applicable in this construction.41


esulting in a complete proposition CLP, which is in exact parallel to PrP or vP withthe involved argument structure fully realized. This is formulated as the <strong>Complex</strong>Classifier Hypothesis (CCH) throughout this thesis:CL and DemP/NumP/AP form a complex classifier, serving as the predicate <strong>of</strong>NP at [Spec,CLP] 7, 8 .In the following section, we will examine the distribution <strong>of</strong> DemP, NumP andAP with CL.2.2.1 Classifiers with Demonstratives<strong>The</strong> basic word order <strong>of</strong> the extended projection <strong>of</strong> nouns in Mandarin isexemplified as follows:7 Num refers to cardinal numerals throughout this work, not to number feature (i.e. singular, dual,plural) or to ordinal numerals.8 Gu Yang (p.c.) points out that Num in the nominal domain is in parallel with Asp in the clausaldomain (Gu, 2005). Since Asp is not a structural building functional element, Num could not be viewedas a structural building functional element. It is well grounded to put NumP in the predicate position.Data from various languages can support this view.(i) How old are you? I am thirty.(ii) What are you? I am a student.In (ii), although ‘a student’ is a DP, it is not an argument, since it does not refer to any individual.Rather it is a property. Since a property is a self-contained element in the sense <strong>of</strong> Chierchia’s semantictheory, we can question the property. Similar reasoning can be extended to (i). NumP serving as matrixpredicate is also seen in other languages like Mandarin.(iii) wo er shi le.我 二 十 了 。I twenty ASP‘I am already twenty (years old).’42


(22) na san jian hong yifu那 三 件 红 衣 服that three CL red coat‘three red coats over there’Mandarin does not have definite article. Definiteness can be expressed in twoways: bare NPs and NPs with demonstratives. However, Mandarin demonstratives arenot English-type demonstratives. English demonstratives are determiners. <strong>The</strong>refore,the co-occurrence <strong>of</strong> definite article and demonstratives is banned.(23) * the that bookMandarin demonstratives are more like Spanish demonstratives. <strong>The</strong>y are APmodifiers. In Spanish (24), Irish (25), and Catalan (26), demonstratives can gotogether with definite article. 9(24) el hombre este (Spanish)the man this‘this man’(25) an leabhar sin (Irish)the book that‘that book’9 <strong>The</strong>se examples are from Lyons (1999: 285).43


(26) els detalls aquells (Catalan)the details those‘those details’According to Lyons (1999), demonstratives locate the entity relative to somereference point in the utterance context. <strong>The</strong> difference between demonstrative anddefinite article is a feature [Dem], which has the value [+Prox]. So this/these isfeatured by [+Def, +Prox], whereas that/those is featured by [+Def, -Prox].A demonstrative signals that the identity <strong>of</strong> the referent is immediately accessibleto the hearer, normally by the speaker’s pointing to the referent. <strong>The</strong> deictic feature[Dem] on demonstratives plays a similar role to pointing, guiding hearer’s attention tothe referent. In other words, the function <strong>of</strong> demonstratives is to facilitate the hearer toidentify the referent. Demonstratives, in this sense, are like relational adverbs hereand there, restricting the exact place. Based on this, I simply take Chinese-typedemonstratives to be APs, in the sense that their semantic function is similar toidentifying adjectives such as identical in the identical car. <strong>The</strong> different distributionand interpretation <strong>of</strong> demonstratives show that in some languages demonstratives areD, and in some languages they are AP.44


In Mandarin, Cantonese, and Yi, demonstratives cannot be used independently. 10<strong>The</strong>y have to go together with classifiers. In English, demonstrative is the head <strong>of</strong> DP;therefore, it can stand alone. In Cantonese and Mandarin, a demonstrative has to gotogether with CL.(27) I bought that (book).10 Gu Yang (p.c.) points out that in Mandarin, demonstratives may be used independently as a pronoun:(i) zhe shi wo de shu.这 是 我 的 书 。this is my book‘This is my book.’(ii) zhe rang wo bu zhidao ru he shi hao.这 让 我 不 知 道 如 何 是 好 。this make me Neg know how be good‘This made me not know what to do.’However, when zhe is used as an independent pronoun, it is used not in the deictic way. It is rather usedas a discourse anaphor, referring to an entity or an event mentioned in the previous utterance. <strong>The</strong>refore,it will be most likely to occur in the construction <strong>of</strong> “zhe shi….” When zhe is used as a real deicticdemonstrative, it is more likely to coocur with a classifier.(iii) zhe shi wode shu.这 是 我 的 书 。this is my book‘This is my book.’(iv) zheben (shu) shi wo de.这 本 ( 书 ) 是 我 的 。this CL (book) is mine‘This book is mine.’(v) wo xihuan zhe ben.我 喜 欢 这 本 。I like this CL‘I like this book.’(vi) *wo xihuan zhe.我 喜 欢 这 。(Intended meaning: I like this book.)We can see that the subject can be zhe/zheben/zhebenshu (this/this CL/this CL book). It is possible forus to assume that when zhe is used as the subject <strong>of</strong> a sentence, it is followed by a silent classifier. Thisview can be further supported by the following data:(vii) zhe xiaoguniang hen ke’ai. ( 这 小 姑 娘 很 可 爱 。)(viii) zhe zhuyi bucuo.( 这 主 意 不 错 。)(ix) zhe ma kuai yao lei si le.这 马 快 要 累 死 了 。this horse soon will tired dead Asp‘This horse is so tired that it is going to die soon.’In Mandarin, it is possible for demonstratives to immediately precede nouns. However, native intuitionshows that there is a silent classifier between the demonstrative and the noun. In Cantonese,demonstratives must be accompanied by the appropriate classifier. It cannot be used as a discourseanaphor.45


(28) ngo5 maai5 zo2 go2 bun2. (Cantonese)我 買 咗 嗰 本I buy ASP that CL‘I bought that (book).’(29) wo mai le na ben. (Mandarin)我 买 了 那 本I buy ASP that CL‘I bought that (book).’<strong>The</strong> difference between English-type demonstratives and Chinese-typedemonstratives can be illustrated as follows:(30)shu ben nabook CL that‘that book’46


In this example, the two noun phrases John and the best candidate are the lexicalconstituents <strong>of</strong> a small clause (cf. Stowell, 1981). Of these two nominals, the formermust be Case-licensed and thus needs to establish a link with the Case-feature <strong>of</strong> theverb consider, which is checked in the checking domain <strong>of</strong> the matrix Agr o (seeChomsky, 1993). Predicate Inversion and SC subject raising are the two ways <strong>of</strong>establishing this link. <strong>The</strong> raising <strong>of</strong> the SC subject is a regular [Spec, PrP]-to-[Spec,Agr o P] Move, since [Spec,PrP] is not a Case-marked position, while [Spec, Agr o P]can check the accusative Case.<strong>The</strong> Case assigning mechanism in Predicate Inversion Move is different from that<strong>of</strong> the regular Spec-to-Spec Move. <strong>The</strong> Case-checking in Predicate Inversion isrealized by the subject-predicate co-indexation (Williams 1980). It is the predicate,rather than the subject, that is raised into [Spec, Agr o P]. <strong>The</strong> subject gets itsCase-feature checked in situ via the chain <strong>of</strong> identical indexations connecting theraised predicate, its trace and the subject <strong>of</strong> the small clause. Viewed in this way,Predicate Inversion is an A-movement process (Den Dikken, 1998) that threatens toviolate the minimality condition, since the raised predicate crosses an interveningA-position (that <strong>of</strong> the small clause subject) on its way up. Such a violation can beavoided if the head whose specifier is crossed in the movement process raises to ahigher head whose specifier functions as the landing site <strong>of</strong> the A-movement processinvolved. That is, the two specifier positions are rendered equidistant from the raisedpredicate’s extraction site. Such a derivation can be illustrated by the following tree:48


(35) wo mai le san ben shu.我 买 了 三 本 书I buy ASP three CL book‘I bought three books.’In this example, the object san ben shu (‘three books’) contains a cardinalnumeral, a classifier and an NP rather than just a cardinal numeral and an NP. <strong>The</strong>word order <strong>of</strong> the nominal is:(36) Cardinal numeral>Classifier>NPIn Yi, classifiers are also obligatory when cardinal numerals go with NPs.However, the word order is different from that <strong>of</strong> Mandarin. <strong>The</strong> word order <strong>of</strong> Yinumeral classifier nominal is:(37) NP>Cardinal numeral>Classifier(38) viex vie nyip buꃧ ꃨ ꑍ ꀮflower two CL‘two flowers’In Jingpo, if the noun is a count noun, the use <strong>of</strong> the classifier is optional. If thenoun is a mass nouns, then the use <strong>of</strong> the classifiers is obligatory. 1313 Examples (39)-(41) are from Cheung (2003: 28).50


(39) sara (marai) masumsă 31 a 33 mă 31 ai 33 mă 31 sum 33teacher CL three‘three teachers’(40) udi (hkum) sanitu 31 ti 31 khum 31 să 31 nit 31egg CL seven‘seven eggs’(41) n-gu *(tum) sanitn 33 -ku 33 tum 33 să 31 nit 31rice CL seven‘seven grains <strong>of</strong> rice’From the above examples, we can see that if Jingpo classifiers appear in thenumeral classifier phrase, the word order <strong>of</strong> the nominal is:(42) NP> Classifier > Cardinal numeral<strong>The</strong> three different word orders can be summarized as follows:51


facts in the structure developed in our proposal. For Mandarin and Cantonese, theclassifier will raise to a higher functional head F, which encodes the feature[+POSS]. 14 <strong>The</strong> cardinal numeral will raise to [Spec, FP], since the raising <strong>of</strong> CL-to-Fhas extended the domain for the head-movement chain containing both [Spec, CLP]and [Spec, FP], which are both equidistant from the extraction site—the complementposition <strong>of</strong> CL. As is argued in Den Dikken (1998), this predicate inversion processnecessitates CL-to-F raising, shown in (44).(44)shu ben sanbook CL threeFor Yi, the word order is NP> Cardinal numeral > Classifier. We assume that thebasic word order is still NP>Classifier>Cardinal numeral. <strong>The</strong> word order observed inYi is derived as follows: (i) move CL to F; (ii) raise CLP to [Spec, FP].14 This [+POSS] feature is responsible for encoding specificity. If the value <strong>of</strong> the feature is [+POSS],the whole nominal expresses possession semantics. <strong>The</strong>refore, the whole nominal will be specific, andan accordance relation will be established between this functional head F and the upper head D, thusthe whole nominal phrase is able to encode definiteness. If the value <strong>of</strong> the feature is [-POSS], thewhole nominal will not express possession semantics, therefore, the whole nominal will be eitherspecific or non-specific. Without adding any other elements, no accordance relation will be establishedbetween F and D, and definiteness cannot be generated with a [-POSS] head.53


2.2.3 Classifiers with AdjectivesLet us now examine the structural relation between classifiers and adjectiveswithin a nominal. In the Yi language, classifiers can also go with adjectives. Likecardinal numerals in Chinese, Yi adjectives cannot go directly with nouns.(45) *co ssakuo la ox.ꊿ ꌳꈄ ꇁ ꀐperson bravecome ASP‘A brave person came.’(46) co ssakuo ma la ox.ꊿ ꌳꈄ ꂷ ꇁ ꀐperson brave CL come ASP‘A brave person came.’<strong>The</strong> nominal structure in (46) can be exemplified by the following tree diagram:(47)co ma ssakuoperson CLbraveHere the classifier will raise up to F, and the remnant CLP will be moved to the[Spec, FP]. F has the [-POSS] feature value, and the classifier adjoins to this54


functional head to jointly encode the specificity <strong>of</strong> the whole nominal phrase. In thiscase, the whole nominal can be either specific or non-specific, which will beeventually determined by the utterance occasion. <strong>The</strong> distribution <strong>of</strong> classifier withadjectives in Yi is different from that in Mandarin and Jingpo, which will beaddressed in the following chapters on reduced relatives in Mandarin and Jingpo.2.3 SummaryChinese-type demonstratives (i.e. demonstratives which are APs rather than D),cardinal numbers and adjectives share a common property. <strong>The</strong>y are potentialpredicates. 15CL is the functional head, which links the argument NP with thepredicate. This is the major syntactic claim <strong>of</strong> the <strong>Complex</strong> Classifier Hypothesis(CCH). As is said before, the basic word order <strong>of</strong> the extended projection <strong>of</strong> nouns inMandarin is (48), exemplified as in (49):15 Gu Yang (p.c.) points out that in Mandarin there are some adjectives that can not be usedpredicatively. For example:(i) zhongxin guangchang ( 中 心 广 场 :central square)(ii) qian zongtong ( 前 总 统 :former president)It is impossible to use the adjectives zhongxin (central) and qian (former) predicatively in Mandarin.(iii) *zhege guangchang hen zhongxin. (* 这 个 广 场 很 中 心 。)(iv) *zhege zongtong hen qian. (* 这 个 总 统 很 前 。)<strong>The</strong> reason why they cannot be used predicatively, I propose, is that these adjectives do not formpredication relation with the nouns guangchang (square) and zongtong (president). Instead they formpredication relation with some silent temporal and spacial nouns. <strong>The</strong>se temporal and spacial nouns,which are licensed by discourse information, cannot surface in Mandarin. <strong>The</strong> structure for (i) and (ii)is as follows:(v) [Silent Noun+ A] NIntuitively, there is an association relation holding between the Silent Noun and the N. <strong>The</strong>refore,semantically it is possible to transfer the property <strong>of</strong> the Silent Noun to the N, although syntactically Ais the predicate <strong>of</strong> the Silent Noun rather than the N. Notice it is not accurate to say that adjectives likezhongxin (central) and qian (former) cannot be used predicately. Rather we would propose that they arepredicates, predicating <strong>of</strong> silent nouns in (i) and (ii). Now we do not know why the temporal andspacial nouns cannot surface. It needs further investigation.Larson (p.c) points out that there is a problem for the above analysis, since we cannot get thepredication: <strong>The</strong> Time is former. Rather, he proposes that there is a generic event operator, working onthe whole event.Apparently, how to deal with those adjectives which can only be used attributively awaits moreresearch in the future. <strong>The</strong>refore, I will not touch upon those adjectives.55


(48) Dem>Numeral>CL>A>N(49) na san jian hong yifu.那 三 件 红 衣 服that three CL red coat‘three red coats over there’<strong>The</strong> basic word order <strong>of</strong> the extended projection <strong>of</strong> nouns in Yi is (50),exemplified as in (51):(50) N>A>Dem>Numeral>CL(51) viex vie a hni a zzyx nyip buꃧꃨ ꀊꅪ ꀊꋨ ꑍ ꀮflower red those two CL‘those two red flowers’As we have discussed before, unlike Bowers, we take AP to be the complement <strong>of</strong>Pr, rather than the adjunct <strong>of</strong> Nm. Thus we stick to the following nominal structure.(52)56


According to the complex classifier hypothesis proposed here, CL (Nm) takesonly one complement. If all the modifiers around the NP appear, then what will be theoutcome? When possessive constructions and relative clauses come into the picture,what will be their positions in the structure? How does the definiteness encodinginteract with the predication formation? <strong>The</strong>se questions will be discussed in thefollowing chapters.57


CHAPTER 3<strong>Complex</strong> Classifiers in Yi and Mandarin3.1 Description <strong>of</strong> Mandarin, Cantonese, and Yi classifiers<strong>The</strong> distribution and interpretation <strong>of</strong> classifiers in nominal structures may varyaccording to different syntactic properties <strong>of</strong> certain functional heads in differentlanguages. In this section, we will focus on the description <strong>of</strong> the distribution <strong>of</strong> theclassifier in Mandarin, Cantonese and Yi. <strong>The</strong> reason why these three languages arestudied is that all <strong>of</strong> them are numeral-classifier languages on the one hand; 1 on theother hand, classifiers are used differently across these three languages.Cheng & Sybesma (1998; 2005) further divides Chinese classifiers intocount-classifier (unit counter) and massifier (measure word; mass-classifier). 2 In thecase <strong>of</strong> mass nouns, counters (measure words, or mass-classifiers) create unit <strong>of</strong>counting; in the case <strong>of</strong> count nouns, counters (count-classifiers) simply name the unitthat the semantic representation <strong>of</strong> the noun provides. Cheng & Sybesma (2005)argues that the classifier is the locus for grammatical number in Chinese. It alsoperforms individualizing function; i.e. it picks out one instance <strong>of</strong> what is denoted byN. Based on these two functions, they generalize a typological parameter that articlelanguages differ from classifier languages in that article languages get definites fromindefinite-turned-definites, while classifier languages derive indefinites on the basis <strong>of</strong>definites.1 According to Allan’s (1977) typology, classifier languages can be divided into numeral classifierlanguages, concordial classifier languages, predicate classifier languages, and intra-locative classifierlanguages. Chinese is a numeral classifier language.2 Chierchia (1998) mentions that saying all members <strong>of</strong> category NP are mass-like does not meansaying that something resembling the mass/count distinction cannot be found in such languages. <strong>The</strong>ycan be registered in the classifier system.58


Returning back to Bowers’ argument that weak determiners are simply APs,which adjoin to Nm’ and have no ability to change the semantic type <strong>of</strong> NmP, we candraw a straightforward conclusion that the [[AP+Cl] + NP] construction should bevery productive, since there should be no constraint on the adjunct AP. However, thisis not the case in Mandarin, as is illustrated in (1).(1) *yi da jian xuexiao [jian: count-classifier]一 大 间 学 校one big CLschool<strong>The</strong> ungrammaticality <strong>of</strong> (1) challenges Bowers’ claim that APs are adjuncts <strong>of</strong>NmP in the case <strong>of</strong> Mandarin. This is also our reason to treat weak determiners andAPs as complements <strong>of</strong> Nm rather than adjuncts throughout this thesis. Recall thefollowing syntactic analyses <strong>of</strong> two men (the left tree is given in Bowers (1991); theright tree is assumed in this thesis).(2)59


However, with massifiers or mensural classifiers, such AP+CL+NP is availablein Mandarin, although the adjectives in this case are restricted to those with thesemantic feature [SIZE].(3) yi da pian di [pian: massifier]一 大 片 地one big CLfiled‘a big piece <strong>of</strong> field’(4) *yi pian da di一 片 大 地one CLbig field(5) yi da bei shui [bei: massifier]一 大 杯 水one big CL water‘a big cup <strong>of</strong> water’(6) *yi bei da shui一 杯 大 水one CL big water60


For massifiers or mensural classifiers, we can use adjectives which encode [SIZE]before classifiers, simply because mass nouns themselves cannot encode SIZE.(7) yi da luo shu [luo: mensural classifier]一 大 摞 书one big CL book‘a big stack <strong>of</strong> books’(8) yi da xiang shu [xiang: mensural classifier]一 大 箱 书one big CL book‘a big box <strong>of</strong> books’(9) *yi hou ben shu一 厚 本 书one thick CLbook(10) yi ben hou shu一 本 厚 书one CL thick book‘a thick book’61


Example (9) is ungrammatical, probably because the classifier ben alreadyencodes SHAPE & SIZE. 3<strong>The</strong>refore, the adjective indicating [SIZE] cannot gotogether with this specific count classifier, while the adjective can go together withthe count noun shu (‘book’), because unlike mass nouns, count nouns as concreteentities with specific boundaries can be further specified by adjectives indicating[SIZE]. What is worth noting here is that in Mandarin it is not possible to have AP-Clprenominals with count nouns, while in Cantonese it is acceptable.(11) *yi da jian xuexiao [Mandarin]一 大 间 学 校one big CLschool(12) yi jian da xuexiao一 间 大 学 校one CLbig school‘a big school’3 Allan (1977: 300) argues that “in oriental languages, books and paper are separately classified. Thispresumably doesn’t happen in other languages for which I have data, because there was no literacyuntil more recently. Many languages also have a ‘general’ or ‘residual’ classifier used <strong>of</strong> a large number<strong>of</strong> heterogeneous inanimate objects, some <strong>of</strong> which may be alternatively classified by more specializedclassifiers.”62


(13) jat1 daai6 gaan1 hok6haau6 [Cantonese]一 大 間 學 校one big CL school‘a school which is big’(14) jat1 gaan1 daai6 hok6haau6一 間 大 學 校one CL big school‘a school which is big’In Mandarin, the adjective indicating [SIZE] has to go together with the countnoun. It is not possible for it to go together with the classifier. However, Cantonesestands in sharp contrast to Mandarin in that the adjective da (‘big’) can go either withthe count noun or with the classifier. <strong>The</strong> semantic focus <strong>of</strong> (13) lies in the ‘bigness’<strong>of</strong> the school; whereas the semantic focus <strong>of</strong> (14) is on the cardinal numeral. <strong>The</strong>syntactic derivation <strong>of</strong> (13) is illustrated in (15); whereas the syntactic derivation <strong>of</strong>(14) is illustrated in (16).63


(15)hok6haau6schoolgaan1jat1onedaai6bigIn this derivational process, the subject <strong>of</strong> the predication relation is hok6haau6( 學 校 ). <strong>The</strong> cardinal predicate and the adjective is complement selected by CL. AfterCL incorporates into F, the complex predicate [NumP+AP] raises to [Spec, FP] viapredicate inversion.(16)hok6daai6haau6daai6hok6haau6gaan1jat164


<strong>The</strong> derivation <strong>of</strong> (14) starts from generating daai6hok6haau6 ( 大 學 校 ‘bigschool’), which involves a direct predication relation and predicate inversion Move.<strong>The</strong>n daai6hok6haau6 ( 大 學 校 ‘big school’) forms a predication relation with thecardinal predicate yi (‘one’). This derivational process captures the intuition that thesemantic focus <strong>of</strong> (14) is on the cardinal predicate rather than the adjective.In (13), both the cardinal numeral and the adjective serve as the predicate <strong>of</strong>hok6haau6 ( 學 校 ). Cantonese data, at the first glance, support Bowers’ claim thatadjectives are adjuncts <strong>of</strong> Nm, because we can generate the following tree for (13) inaccordance with Bowers (1991):(17)jat1 daai6 gaan1 hok6haau6one big CL schoolHowever, this tree cannot capture the predication relation correctly. First, theadjective daai6 ( 大 ) is to predicate <strong>of</strong> the NP hok6haau6 ( 學 校 ). <strong>The</strong> NP is thesubject <strong>of</strong> the predication, and the canonical subject position should be [Spec, NmP].Secondly, treating adjectives as adjuncts cannot capture the semantic differenceshown in (13) and (14). <strong>The</strong>refore, we will not treat adjectives as adjuncts <strong>of</strong> NmP(CLP) in both Mandarin and Cantonese.65


<strong>The</strong> detailed interpretation (referentiality) difference <strong>of</strong> nominals involving theuse <strong>of</strong> classifiers can be seen through such strings as {classifier, NP}, {Possessor,classifier, NP}, and {RC, classifier, NP} in Mandarin, Cantonese and Yi. 4 However,before that we will, first <strong>of</strong> all, examine the distribution and interpretation <strong>of</strong> barenouns in these three languages, which will serve as the foundation for furtherdiscussion.3.1.1 Bare NounsCheng & Sybesma (1999) details the referential property <strong>of</strong> NPs in Cantoneseand Mandarin. <strong>The</strong>ir observation can be summarized as follows:(18)In postverbal position, Mandarin bare nouns can be interpreted as indefinite, definite,or generic. In preverbal position, they can be interpreted as definite or as generic, butnot as indefinite. Cantonese bare nouns cannot be interpreted as definite. Definitecounterparts <strong>of</strong> Mandarin are represented as [CL+N] in Cantonese.Cheng & Sybesma’s explanation <strong>of</strong> the difference between Cantonese andMandarin is that bare nouns in Mandarin are not so-bare. <strong>The</strong>y are CLP. <strong>The</strong> headnoun has been raised to CL and picks up the definite reading from CL, which is theequivalent <strong>of</strong> D in article languages. Bare nouns in Cantonese have not undergone the4 <strong>The</strong> internal word order between the elements in a string {…} is unspecified. For example, the string{classifier, NP} is different from classifier+NP. In the string {Classifier, NP}, the resultant form can beeither classifier+NP or NP+classifier, where “+” means precedence.66


N-to-CL raising. <strong>The</strong>refore, whenever a definite reading is desired, the use <strong>of</strong> CL isobligatory. This difference is illustrated by the following tree diagram:(19)CLPCLPCLNPCLNPNN(Mandarin Bare Nouns)(Cantonese Bare Nouns)Suppose there is a language which encodes definiteness not by CL, but by afunctional head D over CL, then Cheng and Sybesma’s (1999) analysis needs furtherrevision. Yi is such a language. A definite nominal in Yi not only involves the use <strong>of</strong>a classifier, but also a definite article D, which shows that classifiers do notnecessarily exclude the use <strong>of</strong> articles in a language.Before we discuss definite nominals involving CL and D, let us first look at thebare nouns in Yi. Since Yi is an SOV language, with V always occurring at the finalposition, we use the term ‘subject and object’ instead <strong>of</strong> ‘preverbal and postverbal’ toindicate the position <strong>of</strong> bare nouns in the sentence.67


Bare Nouns in subject position:(20) ke vup ddu gat. (Generic)ꈌ ꃷ ꅍ ꇢ .dog bone like-eating‘Dogs like eating bones.’(21) co li kep mu nyi sy yiet ddi. (Generic)ꊿ ꆹ ꈍꃅꑌ ꌦ ꑵꄸ .person TOP anywaydie SFP‘Human beings are mortal anyway.’(22) ke ip nyip ka jjy ka. (Definite)ꈌ ꀃꑍ ꈁ ꐯ ꈁ .dog today happy very happy‘<strong>The</strong> dog is very happy today.’(23) ax yi na ox. (Definite)ꀉ ꑳ ꆅ ꀐ .child ill ASP‘<strong>The</strong> child was ill.’68


(24) co hmat yi go nyi. (Definite)ꊿ ꂘ ꑳ ꇬ ꑌperson classroom insit‘<strong>The</strong> person is sitting in the classroom.’Bare Nouns in object position:(25) nga ke Hxie vur. (Generic)ꉢ ꈌ ꉌ ꃹ.Idog like‘I like dogs.’(26) mu gat tep yy vy bbo o. (Indefinite)ꃅ ꇢ ꄯ ꒉ ꃼ ꁧ ꀑ.Mu Ga book buy go ASP‘Mu Ga has gone to buy books/a book.’(27) mu gat yy ndo sat o. (Indefinite)ꃅ ꇢ ꒉ ꅝ ꌐ ꀑ.Mu Ga soup drink finishASP‘Mu Ga has finished drinking the soup.’69


In subject position, Yi bare nouns can be interpreted as definite, or generic, butnot indefinite. To have an indefinite nominal in that position, there must be a classifier.This is in sharp contrast with Cantonese, for in Cantonese the preverbal CL+NPphrase will be always definite. In object position, Yi bare nouns can be interpreted asdefinite, indefinite or generic. <strong>The</strong> following chart illustrates the distribution andinterpretation <strong>of</strong> bare nouns in Yi, Mandarin, and Cantonese:Table 1: Distribution and interpretation <strong>of</strong> bare nouns in Yi, Mandarin and CantoneseBareIn subject positionIn object positionNouns Definite Indefinite Generic Definite Indefinite GenericYiMandarinxxCantonese x x xFrom this chart, we can see that when bare nouns occur in object position, theycan be indefinite and generic in all the three languages. <strong>The</strong> indefinite bare nounscannot occur in subject positions in all the three languages. This can be accounted forby Longobardi’s (1994) analysis that bare nouns with an indefinite interpretation arerestricted to lexically governed positions (essentially, object position).<strong>The</strong> distribution and interpretation <strong>of</strong> bare nouns in Yi and Mandarin share thesame pattern, which differs from that <strong>of</strong> Cantonese in that bare nouns cannot bedefinite in Cantonese, whereas in Yi and Mandarin they can be definite. Rather than a70


are noun, a [CL+N] string will be used for a definite nominal in Cantonese. Withregard to the use <strong>of</strong> classifiers in Cantonese, Cheng and Sybesma (1999: 518 ) argues:In languages with articles/determiners, the deictic function in the nominalphrase is taken care <strong>of</strong> by the article/determiner. However, this should notlead one to conclude that if a language has no articles/determiners, noelement performs the deictic function. If the describing/referring dichotomyis indeed part <strong>of</strong> Universal Grammar, then if a language has noarticles/determiners, some other element in the language must perform thedeictic function. We suggest that in Chinese Cl performs some <strong>of</strong> thefunctions performed by D, including the deictic function.Cheng and Sybesma contend that bare nouns in Cantonese do not undergo theN-to-CL raising/incorporation. <strong>The</strong>refore, whenever a definite reading is desired, theuse <strong>of</strong> CL is obligatory. In Mandarin, common nouns have undergone the N-to-CLmovement; therefore, bare nouns can have the definite meaning. In order to accountfor the indefinite use <strong>of</strong> bare nouns, Cheng and Sybesma propose that the indefiniteinterpretation <strong>of</strong> nominals in Mandarin is linked to the presence <strong>of</strong> a NumeralP (thehead <strong>of</strong> which may be overt or nonovert), which means that bare nouns not only havean empty CL head, but also an empty Numeral head, and that the differences betweenMandarin and Cantonese can all be traced back to the restriction that in Mandarin, butnot in Cantonese, overt classifiers can occur only with a NumeralP, the head <strong>of</strong> whichmay or may not be overt.Since Yi bare nouns have the same pattern with Mandarin bare nouns, it seems71


that Yi nominals would naturally follow Cheng & Sybesma’s (1999) analysis.However, Yi classifiers complicate the whole picture. We will provide a detaileddiscussion on Yi classifiers in the next section.3.1.2 Yi Classifiers with NP, DemP, AP and NumPYi classifiers can occur with N forming a [NP+CL] string, and such a string canoccur in both the subject and object position. In subject position, it can only havespecific reading, whereas in object position, it can have either specific or non-specificreading.(28) co ma hmat yi go nyi. (Specific)ꊿ ꂷ ꂘ ꑳ ꇬ ꑌ.person CL classroom in sit‘A person is sitting in the classroom.’(29) bbox sse ma a ddit go jjip. (Specific)ꁦ ꌺ ꂷ ꀊꀊ ꇬ ꐛhill CL there at exist‘A hill is there.’72


(30) mu gat tep yy zzit vy bbo o. (Non-specific)ꃅꇢ ꄯ ꒉ ꋐ ꃼ ꁧ ꀑ.Mu Ga book CL buy go ASP‘Mu Ga has gone to buy a book.’(31)ix yi ne lap bbu ji bbop. (Specific)ꀁ ꑳ ꆏ ꇂꇂ ꏢ ꁨ.brother TOP ox CL have‘Brother has an ox.’Now we can compare the distribution and interpretation <strong>of</strong> {CL, NP} string in Yi withthat in Mandarin and Cantonese, which has been detailed in Cheng and Sybesma(1999) in the following table: 55 According to Cheng and Sybesma (1999), the following [CL+N] string in Cantonese can only beinterpreted as non-specific:i) Keoi maai-zo gaa ce.hesell-ZO CL car‘He sold the car.’Different from them, we think [CL+N] in governed (object) position can be both specific andnonspecific. Nonovert Numerals do not always lead to a non-specific reading.According to Cheng & Sybesma (1999), overt classifiers in Mandarin are always accompanied by aNumeral. <strong>The</strong> Numeral can be overt or nonovert. Nonovert Numerals always lead to a non-specificreading due to the lack <strong>of</strong> Quantifier Raising (QR) option.73


Table 2: Distribution and interpretation <strong>of</strong> {CL, NP} in Yi, Mandarin and Cantonese{CL, NP}In subject positionIn object positionDefinite Indefinite Definite IndefiniteSpecific Non-specific Specific Non-specificYi x x xMandarin x x x xCantonese x x<strong>The</strong> crucial difference between Yi and Mandarin/Cantonese is that NP+CL canoccur in subject position and generate specific reading in Yi. In Cantonese CL+NPwill generate definite reading in subject position; whereas in Mandarin CL+NP isbanned to appear in subject position. In Yi, if we want to turn the specific reading intodefinite reading, we can either use a definite article su (ꌠ) as in (33), or usedemonstratives. 6(32) co ma hmat yi go nyi. (Specific)ꊿ ꂷ ꂘ ꑳ ꇬ ꑌ.person CL classroom in sit‘A person is sitting in the classroom.’6 Refer to Liu and Wu (2004) for the detailed illustration <strong>of</strong> how Yi demonstratives encode definitenessin Yi.74


(33) co ma su hmat yi go nyi. (Definite)ꊿ ꂷ ꌠ ꂘ ꑳ ꇬ ꑌ.person CL ART classroom in sit‘<strong>The</strong> person is sitting in the classroom.’Now we can see the three means to encode definiteness in the three languages:Table 3: Subject definiteness encoding in Mandarin, Yi and CantoneseDefiniteness PatternMandarinSubject Position NoteBare NPYiCantoneseBare NPNP+CL+suCL+NPNP+CL (specific indefinite)With regard to Yi, there are two ways to encode definiteness. <strong>The</strong> first way is thesame as seen in Mandarin; that is, through bare nouns. <strong>The</strong> second way is through theinsertion <strong>of</strong> classifiers and the definite article -su. For bare nouns, we assume thatN-to-D movement occurs. <strong>The</strong> following tree illustrates this movement:(34)75


For bare nouns, there exists no CL projection. D will directly take an NP as itscomplement. <strong>The</strong> Head <strong>of</strong> NP will get raised to the head <strong>of</strong> DP via Head-to-Headagreement. Bare nouns in this case becomes DPs and automatically get the referringfunction. This explains the Mandarin and Yi cases.If a definite nominal contains CL, two options arise. <strong>The</strong> first one, represented byCantonese, is to assign the deictic function totally to CL; thus Cheng and Sybesma’s(1999) analysis. <strong>The</strong> second option, represented by Yi and Mandarin, is to assign thedeictic function partially to CL. <strong>The</strong> referential property cannot be fully taken by CL.Instead it has to resort to some other elements. For Mandarin, a demonstrative isrequired to encode the definiteness; whereas for Yi, besides using demonstratives, Yican also use a definite article. <strong>The</strong> reason why a definite article is required is due tothe Head Movement Constraint (HMC), proposed by Travis (1984). HMC bars a headcrossing another head in the movement process. N-to-D movement cannot take place,since there is an intervening head CL. <strong>The</strong>refore, instead <strong>of</strong> taking the strategy <strong>of</strong>Move, Yi takes the strategy <strong>of</strong> Merge. CLP will be merged with a definite article toencode definiteness. In a nutshell, in classifier languages either CL raises to D tocheck the D feature, or a definite article merges to D to check the D feature.Besides the definite article, Yi classifiers can also go together with other elements.First, CL can go together with adjectives.76


(35) nga ax nuo ji vy mix.ꉢ ꀉ ꆈ ꏢ ꃼ ꂯI black CL buy ASP‘I will buy a black one.’(36) co ssakuo ma la ox.ꊿ ꌳꌳ ꂷ ꇁ ꀐperson brave CL come ASP‘A brave person came.’Secondly, CL can go together with demonstratives as well as with cardinal numerals.(37) ax yi cyx ma i dix nyip ggu jjo.ꀉꑳ ꋋ ꂷ ꀂ ꄁ ꑍ ꈬ ꐥchild this CL coat two CL have‘This child has two coats.’Yi nouns cannot cooccur with numerals or demonstratives unless classifiers are used.(38) *ax yi cyx i dix nyip ggu jjo.ꀉꑳ ꋋ ꀂ ꄁ ꑍ ꈬ ꐥchild this coat two CL haveIntended meaning: ‘This child has two coats.’77


(39) *ax yi cyx ma i dix nyip jjo.ꀉꑳ ꋋ ꂷ ꀂ ꄁ ꑍ ꐥchild this CL coat two have‘This child has two coats.’When demonstrative, numeral, classifier, adjective all occur in a nominalexpression, there is a strict word order: N>A>Dem>Num>CL. For example:(40) ngat Nuosuvitgga ashyt cyx nyip ggu bipji lur kur da ggut su nge.ꉠ ꆈꌠ ꃢꃢ ꀊꀊ ꋋ ꑍ ꈬ ꀙꏢ ꇓꇓ ꄉ ꈪ ꌠ ꉬmy Yi coat new this two CL Beijing city in sew SU be‘<strong>The</strong>se two new Yi coats <strong>of</strong> mine were made in Beiing.’ (Chen & Wu: 76)(41) nga viex vie a hni a zzyx nyip bu vy six la ox.ꉢ ꃧꃧ ꀊꀊ ꀊꀊ ꑍ ꀮ ꃼ ꌊ ꇁ ꀐI flower red that two CL buy PRT come ASP‘I bought those two red flowers.’ (Chen & Wu: 76)From the above examples we can see that AP>DemP>NumP are between thehead noun and the classifier, where each <strong>of</strong> AP, DemP and NumP can occurindependently. We would argue that the complex predicate [AP+DemP+NumP]occupies the same position as the simple predicate AP, DemP or NumP. <strong>The</strong>hierarchy between them can be readily explained by Cinque (2005), which will be78


discussed in 3.3.3.1.3 Classifiers with possessor NPs and relative clausesIn terms <strong>of</strong> complex nominals which involve possession semantics or relativeclauses, Cantonese is quite unique. Cantonese possesses definite structures such asCL-NP, Poss-CL-NP and RC-CL-NP phrases that Mandarin lacks. We can observethat Cantonese has the sequence XP-CL-NP where XP can be a null element, apossessor noun phrase and a relative clause, and from the semantic point <strong>of</strong> view, thisstructure XP-CL-N denotes definiteness. <strong>The</strong> syntactic difference between Cantoneseand Mandarin classifiers is exemplified as follows: 7(42) bun2 syu1 m4gin3 zo2 (Cantonese)本 書 唔 見 咗 。CL book not see ASP‘<strong>The</strong> book has disappeared.’(43) siuming bun2 syu1 m4gin3 zo2小 明 本 書 唔 見 咗 。Siuming CL book not see ASP‘That book <strong>of</strong> Siuming’s has disappeared.’7 Examples (42) to (47) are from Au Yeung (2005) with minor modifications in glossing.79


(44) dit3 zo2 lok6 heoi3 bun2 syu1 m4 gin3 zo2跌 咗 落 去 本 書 唔 見 咗 。Fall ASP fall go CL book not see ASP‘<strong>The</strong> book that has fallen down has disappeared.’According to Au Yeung (2005), Mandarin classifiers do not have [+d] value, but[-d] as its default value; 8 therefore, it does not form definite CL-N phrases (45). Nordoes it have Poss-CL-N or RC-CL-N sequence, as in (46) and (47).(45) * ben shu bu jian le. (Mandarin)本 书 不 见 了 。CL book not see ASPIntended meaning: ‘<strong>The</strong> book has disappeared.’(46) * Xiaoming ben shu bu jian le.小 明 本 书 不 见 了 。Xiaoming CL book not see ASPIntended meaning: ‘That book <strong>of</strong> Xiaoming’s has disappeared.’8 <strong>The</strong> feature [+d] <strong>of</strong> classifiers refers to the ability to encode definiteness. [+d] means classifiers havethe ability to encode definiteness. [-d] means classifiers have no ability to assign the definitenessfeature to a NP.80


(47) * Diao le xia qu ben shu bu jian le.掉 了 下 去 本 书 不 见 了 。fall ASP down go CL book not see ASPIntended meaning: ‘<strong>The</strong> book that has fallen down has disappeared.’In order to express possessive constructions and relative clauses, the morpheme –dehas to be used. When –de is used, demonstrative zhe/na (‘this/that’) will come into thesentence naturally to encode definiteness.(48) Xiaoming de na ben shu bu jian le.小 明 的 那 本 书 不 见 了 。Xiaoming DE that CL book not see ASP‘That book <strong>of</strong> Xiaoming’s has disappeared.’(49) Diao le xia qu de na ben shu bu jian le.掉 了 下 去 的 那 本 书 不 见 了 。fall ASP down go DE that CL book not see ASP‘<strong>The</strong> book that has fallen down has disappeared.’ways:In Yi, the semantic relation <strong>of</strong> possession can be expressed in the following three81


(50) a. Possessor NP1+ Possessee NP2b. Possessor NP1+ Possessee NP2 +(Num) CL-øc. Possessor NP1+Possessee NP2 +(Num) CL-su<strong>The</strong> first way is simple. We just put the possessor NP before the possessee NP, andthis two NPs structure will automatically generate the possession semantics. However,notice in Yi when two NPs are put together (NP1+NP2), there are actually twointerpretations: possession and apposition, exemplified in (51) and (52) respectively.(51) mu gat hmatmopꃅꇢMugaꂘꂘteacher‘Muga’s teacher(s)’(52) hmatmop mu gatꂘꂘteacherꃅꇢMuga‘Teacher Muga’This structural ambiguity can be easily identified according to the types <strong>of</strong> theNPs involved. (51) is the typical possessive construction in Yi. Both the possessor NPand the possessee NP are referential NP, bearing the possessor and possessee thetarole. <strong>The</strong>y are the arguments <strong>of</strong> an abstract POSSESS predicate, which establishes thepossession relation. (52) is a typical appositive structure, in which hmatmop (‘teacher’)82


is the title <strong>of</strong> the person named Muga. <strong>The</strong>re is presupposition for this appositivephrase: Muga is a teacher. <strong>The</strong> first NP (NP1) in appositive phrase is a nominalpredicate. <strong>The</strong>re is a predication relation holding between NP1 and NP2. <strong>The</strong> surfaceword order is derived by predicate inversion. <strong>The</strong>refore, unlike NPs in possessives,the first NP in appositive phrase is a nominal predicate, whereas the second NP (mostlikely a DP) is its argument.When specifying the number <strong>of</strong> the possessee (the possessive head), Yi resorts tonumerals and classifiers. When the number is one, the cardinal numeral one can beomitted, as in (53).(53) mu gat ax yi maꃅꇢ ꀉꑳ ꂷMugachild CL‘a child <strong>of</strong> Muga’s’(54) mu gat ax yi nyip maꃅꇢ ꀉꑳ ꑍ ꂷMuga child two CL‘Muga’s two children’(54) is a specific indefinite nominal. It presupposes the existence <strong>of</strong> two children.Suppose Muga has five children. Two <strong>of</strong> them are at home, and three <strong>of</strong> them are notat home. (54) is an example <strong>of</strong> partitive phrases. By definition, partitives are specific,but not definite. In order to turn it into a definite phrase, the definite article su will be83


used. That involves the third way to encode possession in Yi.(55) mu gat ax yi nyip ma suꃅꇢ ꀉꑳ ꑍ ꂷ ꌠMuga child two CL ART‘the two children <strong>of</strong> Muga’s’This phrase presupposes the existence <strong>of</strong> two children, and the reference to thetwo children has to be applied to the entirety. That is to say, the set denoted by twochildren is an inseparable unit. That is exactly where definiteness hinges upon.<strong>The</strong> description for Yi possessives can also be extended to Yi relative clauses.Like Cantonese, Yi relatives can go with CL, but unlike Cantonese, if the relativeclause goes with a classifier, the head noun (the relative head) is not definite; rather itis indefinite. In order to have a definite relative head, Yi uses the definite article su.What is most interesting is that RC can also directly go with su without classifiers. Inthis case, the number feature <strong>of</strong> the relative head is unspecified. It can be eithersingular or plural. Relevant examples are shown in (56)-(58): 9(56) vot [mu gat hxo ta] ma at guop gep sit ox.ꃮ ꃅ ꇢ ꉘ ꄉ ꂷ ꀈꇩ ꇱ ꌉ ꀐpig Muka feed ASP CL Aguo AGENT kill ASP‘Aguo killed a pig that Mu Ka fed.’9 Relative clause is marked by square bracket for ease <strong>of</strong> exposition.84


(57) vot [mu gat hxo ta] ma su at guop gep sit ox.ꃮ ꃅ ꇢ ꉘ ꄉ ꂷ ꌠ ꀈꇩ ꇱ ꌉ ꀐpig Muka feed ASP CL ART Aguo AGENT kill ASP‘Aguo killed the pig that Mu Ka fed.’(58) vot [mu gat hxo ta] su at guop gep sit ox.ꃮ ꃅ ꇢ ꉘ ꄉ ꌠ ꀈꇩ ꇱ ꌉ ꀐpig Muka feed ASP ART Aguo AGENT kill ASP‘Aguo killed a pig/pigs that Mu Ka fed.’3.2 Previous analyses <strong>of</strong> Classifier PhraseWe find that no theories on classifiers, or nominal structure <strong>of</strong> classifierlanguages, currently available can readily account for the Yi data without anyproblem. In the following I will review three influential proposals on classifierprojection.3.2.1 Tang (1990); Cheng and Sybesma (1999)Cheng and Sybesma (1999) adopts Tang’s (1990) proposal for the structure <strong>of</strong>noun phrase, which is shown in (59). <strong>The</strong> main character <strong>of</strong> this kind <strong>of</strong> analysis isthat there is a NumeralP between DP and CLP. Cardinal numerals occupy the headposition <strong>of</strong> Numeral. Demonstratives are D.85


(59)<strong>The</strong> basic observation, which Cheng and Sybesma (1999) consider as the startingpoint <strong>of</strong> their work, is that Chinese languages allow bare nouns to appear asarguments. According to the claim made by Stowell (1989), Szabolcsi (1994),Longobardi (1994), and others that only DPs can function as arguments, bare nouns inChinese must involve more structure than just the bare N (or the bare NP). Bare nounsmust have a functional projection which is headed by a D. A question that naturallyarises here is what will be the instantiation <strong>of</strong> this functional category D in Chinese.Although this DP claim is well grounded in case <strong>of</strong> non-classifier languages likeEnglish and Italian, there remains a question whether it can be readily applied toclassifier languages like Chinese. Chierchia (1995,1998) proposes that Chinese nounsdiffer from nouns in languages like English in that they are arguments (<strong>of</strong> type )rather than predicates (<strong>of</strong> type ) and thus can appear bare. In this view, there isno need for bare nouns to project further into higher configuration. In order todetermine which view appropriately reflects the Chinese grammar, Cheng and86


Sybesma conduct a detailed study on the distribution and interpretation <strong>of</strong> Chinesenominal structures.<strong>The</strong> fact that definite bare nouns can occur preverbally indicates that the head <strong>of</strong>the CLP is not empty. 10 <strong>The</strong>refore, the authors propose that in cases where a barenoun receives a definite interpretation, N has moved to CL: once the CL position hasbeen filled, the phrase is no longer limited to occurring in lexically governed positions.<strong>The</strong> source for the definite interpretation is argued to be the non-overt ι operator. 11Cheng and Sybesma further argue that the N-to-Cl is a necessary step for use <strong>of</strong> the ιoperator either because the ι operator changes the NP into an individual ,which cannot stay in an NP position and thus must undergo movement, or because theN must be in CL position for the ι operator to function.Indefinite noun phrases <strong>of</strong>ten bear the form [Cl+N] or [yi-Cl+N]. <strong>The</strong>re is a viewthat [Cl+N] derives from [yi-Cl+N] by phonological reduction. <strong>The</strong> authors argueagainst this view via two contexts where only [Cl+N] (but not [yi-Cl+N]) can survive.<strong>The</strong> first context involves bounded predicates (typically, verbs withaccomplishment situation aspect). <strong>The</strong> object <strong>of</strong> the bounded predicate cannot benon-specific noun phrases. <strong>The</strong>refore, [CL+N] cannot appear in the object position <strong>of</strong>the bounded predicates, whereas [yi-CL+N] can have a specific reading. <strong>The</strong>Mandarin ba-construction is another context <strong>of</strong> boundedness. <strong>The</strong> NPs following ba10 If the head <strong>of</strong> ClP is empty, there are two implications. <strong>The</strong> first is that the empty head is occupied byan existential operator, following Longobardi’s (1994) analysis. <strong>The</strong>n the bare noun phrase cannot bedefinite. <strong>The</strong> second is that the empty operator, being an empty category, must be properly governed, inthis case, lexically governed. That is, the bare NP must appear in the post-verbal position so that theempty operator can be lexically governed by the V.11 Chierchia (1998) assumes that the operator is a type-shifter and that it is equivalent to a definitearticle.87


cannot be non-specific. <strong>The</strong>refore, [CL+N] phrases cannot occur in that position.Based on these observations, the authors conclude that [CL+N] phrases in Mandarinmust be nonspecific indefinites.<strong>The</strong> second context involves sentences with secondary predicates. <strong>The</strong> objectNPs in these sentences must be specific indefinites, as is discussed in Huang (1987).<strong>The</strong>refore, [CL+N] phrases cannot appear as the object in sentences involvingsecondary predicates. <strong>The</strong> following two sentences show a clear difference between[CL+N] phrases and [yi-CL+N] phrases:(60) wo jiao guo yi ge xuesheng hen congming.我 教 过 一 个 学 生 很 聪 明 。I teach EXP one CL student very smart‘I once taught a student who is very smart.’(61) *wo jiao guo ge xuesheng hen congming.我 教 过 个 学 生 很 聪 明 。I teach EXP CL student very smartBased on the discussion <strong>of</strong> the two contexts, the authors conclude that thedifferent distribution pattern <strong>of</strong> these two types <strong>of</strong> phrases show that in Mandarin[CL+N] and [yi-CL+N] differ in that [CL+N] can only have non-specific indefiniteinterpretation, whereas [yi-Cl+N] can have both specific and non-specificinterpretation.88


Under Cheng and Sybesma’s analysis, if bare noun phrases are definite, they areCLPs, and their definiteness comes from the ι operator, which can be considered to belike a definite article. <strong>The</strong> numeral apparently has the effect <strong>of</strong> UNDOING thedefiniteness. Indefinite bare noun phrases and indefinite [CL+N] phrases are bothNumeralPs, the former having an empty Numeral head and an empty CL head and thelatter only an empty Numeral head. This leads the authors to conclude that an emptyNumeral head leads to a nonspecific interpretation, and the authors use this to explainwhy [yi-CL+N] can be interpreted as both specific and nonspecific, whereas indefinitebare nouns and [CL+N] can only be interpreted as nonspecific. 12Cheng and Sybesma illustrate the structure <strong>of</strong> definite and indefinite noun phrasesby the following trees:(62)12 An alternative method <strong>of</strong> dealing with specificity is provided in Givon (2001), which I believe istrue. Givon noticed that Quine (1953) was the first to note the systematic connection between thereference status <strong>of</strong> nominals and the propositional modalities. Consider the following examples:a. John married a rich woman.b. John wanted to marry a rich woman.i. …but she refused him.ii. …but he couldn’t find any.<strong>The</strong> speaker uttering sentence (a) is committed to the existence <strong>of</strong> a rich woman that John married inthe universe <strong>of</strong> discourse. However, the speaker in uttering sentence (b) may or may not have madesuch a referential commitment. <strong>The</strong>refore, sentence (b) has a referring interpretation as in (bi) and anon-referring interpretation as in (bii). <strong>The</strong> indefinite article fails to determine the referential property<strong>of</strong> the nominal expressions. Rather, it is the propositional modality that makes such a determination. Inthe modality <strong>of</strong> realis, nominals can only be interpreted as referring, whereas in the irrealis modality,nominals can be interpreted as either referring or non-referring.89


NumeralPs (due to the existential quantificational force) are indefinite, whereasCLPs (either with an overt CL or with the ι operator and the N-to-Cl movement) aredefinite. What is different between Mandarin and Cantonese is that in Mandarin, butnot in Cantonese, overt classifiers can occur only with a NumeralP, the head <strong>of</strong> whichmay or may not be overt. This means in Mandarin whenever a classifier occurs, thewhole nominal expression must be a NumeralP, thus must be indefinite.Cheng and Sybesma’s analysis captures the interpretative difference betweenMandarin classifier phrase and Cantonese classifier phrase. However, this analysiscannot be readily extended to other Sino-Tibetan classifier languages. For example, inYi a definite noun phrases will involve the use <strong>of</strong> both the definite article and aclassifier. A definite noun phrase in Jingpo may involve the use <strong>of</strong> the definitearticle. 133.2.2 Li (1999)According to Li’s (1999) analysis, there is a NumP between DP and CLP. <strong>The</strong>feature carried by the head Num is [+ plural]. <strong>The</strong> plurality feature is grammaticalizedas the Head Num. <strong>The</strong> [Spec, NumP] position is the position for cardinal numerals.Li’s analysis is different from Tang’s in the treatment <strong>of</strong> status <strong>of</strong> cardinal numerals.In Li’s analysis, they are XPs occupying specifier positions; whereas in Tang’sanalysis, they are heads <strong>of</strong> NumeralPs.13 For count nouns in Jingpo, the use <strong>of</strong> classifiers is even optional. That is, classifiers play no role indetermining the definiteness in this language.90


(63)santhreegeCLxueshengstudentDifferent from Chierchia’s (1998) claim that nouns in classifier languages aremass nouns, and that mass nouns, being inherently plural, do not have pluralmorphology, Li argues that a classifier language can have a plural morpheme within anominal expression. Since classifier languages differ from non-classifier languages inthat the classifier projection does not occur in the latter, the syntactic realization <strong>of</strong>plurality in these two types <strong>of</strong> languages differ in that the plural morpheme is realizedin N in non-classifier languages, whereas it is realized in D in classifier languages.This point is <strong>of</strong> great significance not only because it explains the asymmetry <strong>of</strong>plurality realization in a typology <strong>of</strong> classifier languages and non-classifier languages,but also because it further elucidates the DP structure and gives strong support to theDP hypothesis for classifier languages.Li argues that although Chinese does not have an equivalent <strong>of</strong> the definitearticle the in English, it still has a DP structure for argument nominal expressions, asis shown above in (63). Pronouns and proper names, being definite expressions91


inherently, can be base-generated in D. Common nouns, however, can be interpretedas definite if they are raised from N to D. <strong>The</strong> N-to-D raising is not possible whenthere is an intervening head between N and D, such as Num or CL (the HeadMovement Constraint (HMC) forbidding a head crossing another head in themovement process). If –men is regarded as the head <strong>of</strong> NumP, then it should bec-commanded by D. <strong>The</strong>refore, pronouns and proper names can go before –men. Liargues that common nouns are base-generated in N, but there are cases wherecommon nouns are in D, as a result <strong>of</strong> movement from N to D, being interpreted asdefinite. When a common noun is in D, it can be suffixed with –men. This explainsthe phenomenon that common nouns with–men must be definite. Since -men issuffixed to a common noun only when the noun is moved up to D, a common nounstaying in the N position cannot be suffixed with -men. A quantity expression[Num+CL] therefore cannot precede N-men, according to the phrase structure [D[Num [CL [N]]]].In the same spirit, Li proposes the following structure for the English pluralnominal phrase them three:(64)92


Here the pronoun is base-generated in D. If the head <strong>of</strong> a nominal phrase (e.g.three students) is a common noun with student being base-generated at N, the pluralmorpheme –s will be base-generated in Num. student can move up to Num to realizethe plural feature.Li argues that Chinese, being a Classifier language, has an additional Classifierprojection between the Number projection and N:(65)santhreegeCLxueshengstudentXuesheng (‘student’) cannot move up to Num for the Pl feature to be realized,because the head noun cannot move across another head CL. Li argues that the onlyoption left is for the Pl feature to be raised to D and suffixed to the nominal element inD. <strong>The</strong>refore, it is grammatical to get phrases like tamen san ge xuesheng (‘they threeCL students’). Li further argues that this account also predicts that when there is nointervening Cl, the plural morpheme –men can be realized in N:(66) a. laoshi dui zhexie/naxie xuesheng-men tebie hao.老 师 对 这 些 / 那 些 学 生 们 特 别 好 。teacher to these/those student-Pl especially good‘<strong>The</strong> teacher is especially nice to these students.’93


. *laoshi dui zhe/na ji ge xuesheng-men tebie hao.老 师 对 这 / 那 几 个 学 生 们 特 别 好 。teacher to this/that several CL student-Pl especially goodIntended: ‘<strong>The</strong> teacher is especially nice to these/those several students.’<strong>The</strong> lack <strong>of</strong> a classifier makes it possible for the suffix -men to be attached to thenoun. Li summarizes the behavior <strong>of</strong> the Chinese plural morpheme -men by sayingthat plurality is specified in the head <strong>of</strong> the Number projection in both English andChinese, and that English has an obligatory agreement relation between Number andNoun with the noun obligatorily raised to Num, and with the plurality feature realizedon N; whereas in Chinese, an N cannot be raised at all to Num when there is aClassifier projection between Num and N; therefore the plural morpheme in Chineseis generally suffixed to an element in D.Li’s analysis differs from Cheng & Sybesma’s in that Li argues for the existence<strong>of</strong> D, which is the host <strong>of</strong> the plural morpheme in nominal structure. Under Li’sanalysis, D can only be occupied by nominal elements such as proper names, personalpronouns and common nouns which have undergone N-to-D movement. This isunderstandable because there is no definite article in Mandarin. In some otherclassifier languages such as Jingpo, they do have the definite article and the pluralmorpheme. However, it is impossible to affix the plural morpheme to the definitearticle. We will discuss this issue in Chapter 6.94


3.2.3 Shen, He, and Gu (2001); Au Yeung (2005)In Shen, He, and Gu (2001: 634), cardinal numerals are treated as QPs, and basegenerated at [Spec, CLP]. <strong>The</strong>re is no intervening projection between DP and CLP.This is also the position taken by Au Yeung (2005), which he dubs asNumP-in-SpecCLP model.(67)DPD’one CL bookAccording to this view, the nominal structure <strong>of</strong> both Cantonese and Mandarin isassumed to have the CLP model, where the head CL and NP are in head-complementrelation, NumP is in the specifier <strong>of</strong> CLP, and the demonstrative is located in [Spec,NumP]. CL heads its own projection and complement-selects an NP. Au Yeung (2005)argues that this CLP structure can predict not only the word order <strong>of</strong> the typicalMandarin data studied in the literature, but also that <strong>of</strong> possessive and relativizednominals in Cantonese. Since this model is exactly XP-CL-NP model, differentelements can be well placed in this model. This XP can be either DemonstrativePhrase, Possessor (DP), or RC.95


As for the definiteness feature <strong>of</strong> the functional head CL in Cantonese, thedefinite value [+d] and indefinite value [-d] can be checked <strong>of</strong>f by matching the value<strong>of</strong> the element in [Spec, CLP]. When [Spec, CLP] is empty, it is assumed to have thedefinite value as default, and check <strong>of</strong>f the definite value [+d] <strong>of</strong> the definitenessfeature in the functional head CL. When [Spec, CLP] is occupied by NumP, which isinterpreted as indefinite, the NumP is assumed to have [-d] value, which then checks<strong>of</strong>f the [-d] value <strong>of</strong> the definiteness feature in the functional head CL.This model can easily explain why Cantonese possessives and relatives can bemarked by classifiers; whereas Mandarin possessives and relatives cannot. However,this model shares the same problem with Cheng and Sybesma’s analysis that it failsto notice the fact that in many classifier languages the definiteness encoding is not(purely) taken by classifiers.3.2.4 Interium summaryAlthough different proposals argue for different syntactic structure <strong>of</strong> Chinesenominals, namely, the Classifier Phrase in Cheng & Sybesma’s model, the DeterminerPhrase in Li’s model, and the Qantifier Phrase in Au Yeung’s model, they share thesame idea that the legitimacy <strong>of</strong> definite Classifier-Noun phrases in Cantonese but notin Mandarin lies in the optionality <strong>of</strong> Numeral Phrase in the nominal structure. InMandarin, Numeral Phrase has to be obligatory in the presence <strong>of</strong> the classifier andhence there is no way for Classifier-Noun phrases to denote definiteness. What is leftfor Mandarin to denote definiteness is either Demonstrative-Numeral-Classifier-Noun96


phrases or bare nouns. In Cantonese, Numeral Phrase is optional in the nominalstructure even if the classifier is present and therefore Classifier-Noun phrases(without any numeral) can be used as definite.Most importantly, all <strong>of</strong> the three analyses reviewed above share a similarity thatunder each analysis CL complement-selects NP. <strong>The</strong>y all treat the NP as thecomplement <strong>of</strong> the CL. <strong>The</strong> problem with this kind <strong>of</strong> analysis is that it fails tocapture the close parallelism between nominal structure and clausal structure in terms<strong>of</strong> the predication relation. Different from their analyses, this thesis has proposed analternative for the nominal structure in Sino-Tibetan classifier languages, i.e. the<strong>Complex</strong> Classifier Hypothesis (CCH), repeated here as follows:(68)This alternative analysis is more advantageous than the analyses just reviewed inthat it can capture the predication relation holding between the NP and the predicatessuch as AP, DemP and NumP. <strong>The</strong> CCH takes the view that in classifier languages,nouns are mapped to arguments. <strong>The</strong>y are in the specifier position, CL is thepredication operator, as is argued in Bowers (1991), and the predicates (AP, DemPand NumP) are the complement <strong>of</strong> CLP. However, there is a strict hierarchy betweenthese predicates when all <strong>of</strong> them occur in a single nominal. <strong>The</strong> universal hierarchyis proposed in Cinque (2005). We will take a closer look at it in the following.97


3.3 Cinque’s Universal Order <strong>of</strong> NP ModifiersCinque (2005) examines Greenberg’s Universal 20 (1963) and its exceptions, 14and argues that Universal 20 is derived from a universal hierarchy for modifiersaround the NP. <strong>The</strong> universal hierarchy is Dem>Num>A>N. All other orders, bothpronominal and postnominal, attested in different languages are derived by moving(or not moving) the NP around the modifiers, which are base-generated prenominallyin the fixed order Dem>Num>A. If nothing moves, the unique order foundprenominally (the Dem>Num>A>N order) surfaces. As for the postnominal orders,they arise via the two ways in which the NP raises: either alone, from specifier tospecifier (Spec-to-Spec) <strong>of</strong> Agr(eement) projections above each <strong>of</strong> the functionalprojections hosting adjectives, numerals, and demonstratives, to give the order N>Dem>Num>A, or by moving successively to each such Spec and pied-piping thecategory that dominates it, in a “roll-up” fashion that reverses the order <strong>of</strong> themodifiers, to give N>A>Num>Dem.14 “When any or all <strong>of</strong> the items (demonstrative, numeral, and descriptive adjective) precede the noun,they are always found in that order. If they follow, the order is either the same or its exact opposite” (p.87)98


(69)Although Cinque’s hierarchy is purely based on the observation <strong>of</strong> syntacticdistribution <strong>of</strong> different elements within a nominal phrase across a large number <strong>of</strong>languages, I believe there is semantic justification for the underlying universalhierarchy Dem>Num>A>N. This hierarchy reflects the wrapping <strong>of</strong> a bare noun withdifferent coating agents until it can refer. Each layer <strong>of</strong> coating restricts thedenotations <strong>of</strong> the bare noun step by step. For example: apples, big apples, three bigapples, those three big apples. We can see that the denotation <strong>of</strong> the nouns phrase bigapples is a subset <strong>of</strong> the denotation <strong>of</strong> the noun apples. <strong>The</strong> denotation <strong>of</strong> three bigapples is a subset <strong>of</strong> that <strong>of</strong> big apples. <strong>The</strong> denotation <strong>of</strong> those three big apples is asubset <strong>of</strong> that <strong>of</strong> three big apples.Building in parallel the <strong>Complex</strong> Classifier Hypothesis and Cinque’s insightabout the relative order between Dem, Num, and A, we can get a fuller illustration <strong>of</strong>99


nominal structure:(70)Cinque’s universal hierarchy treats DemP, NumP and AP as modifiers <strong>of</strong> NP.Under our CCH, DemP, NumP and AP are predicates, predicating <strong>of</strong> the subject at[Spec, CLP]. That is, under CCH, there is no modification relation. Everymodification is a predication relation, as long as there is a predicate. Figurativelyspeaking, the predication relation can be viewed as the unwrapping <strong>of</strong> the coating.Modification is from abstract to concrete, whereas predication is from concrete toabstract. Despite the difference between modification and predication, the relativeword order between DemP, NumP and AP, we propose, still follows Cinque’suniversal hierarchy for both syntactic and semantic reasons. <strong>The</strong> syntactic reason isgiven by Cinque that that word order is widely attested cross-linguistically. <strong>The</strong>semantic reason is that people prefer to predicate <strong>of</strong> an entity from near to far, fromconcrete to abstract.100


3.4 Extended projection <strong>of</strong> nounsWith the CCH in mind, we are going to check it against data from both Yi andMandarin and see whether the CCH holds for both languages and show how the CCHdetermines the definiteness <strong>of</strong> noun phrases in Yi and Mandarin.3.4.1 Extended projection <strong>of</strong> indefinite nouns in YiFrom the previous discussion, we know that DemP, NumP and AP can all serveas the complement <strong>of</strong> CL in Yi. However, when they co-occur, there is a strict wordorder requirement. Demonstratives are special in that they interact with definite article-su. A definite nominal in Yi may have three forms: bare noun, N+Dem+CL, andN+CL+su. Whenever a demonstrative is used, the definite article cannot cooccur.<strong>The</strong>refore, in indefinite nominal we can only have Numeral and Adjectives. <strong>The</strong> basicword order <strong>of</strong> the extended projection <strong>of</strong> indefinite noun phrase in Yi is N>A>Numeral>CL, as is exemplified in (71):(71) viex vie a hni nyip buꃧꃧ ꀊꀊ ꑍ ꀮflower red two CL‘two red flowers’When the DP is indefinite, D has no phonetic realization. CL will adjoin to thisempty D with the feature [-def] via F (the head <strong>of</strong> (N+I)P or POSS in Szabolcsi’s(1994) sense). <strong>The</strong>refore the indefinite nominals are marked by [CL+F+ø]. After CL101


has incorporated into D, the remnant CLP moves to [Spec, DP], resulting in thedesired indefinite nominal string [NP+AP+NumP+CL]. <strong>The</strong> derivation is shown asfollows: 15(72)15 This analysis follows from the derivation <strong>of</strong> the sentential SOV word order in Yi. Assuming thatSpecifier-Head-Complement is the universal word order (Kayne 1994), we can derive SOV via thefollowing derivation:102


<strong>The</strong> surface word order <strong>of</strong> AP and NumP is derived in the spirit <strong>of</strong> Cinque (2005).(73)3.4.2 Extended projection <strong>of</strong> definite nouns in YiFrom the previous discussion, we know that there are two basic word orders <strong>of</strong>the extended projection <strong>of</strong> definite noun phrase in Yi corresponding to two differentways to encode definiteness (c.f. Liu and Wu 2004). <strong>The</strong> first way is to resort todefinite article su, as is illustrated in (74). <strong>The</strong> second way is to resort todemonstratives, as is illustrated in (75). We will examine the syntactic derivation <strong>of</strong>them in the following sections.(74) viex vie a hni nyip bu suꃧꃧ ꀊꀊ ꑍ ꀮ ꌠflower red two CL ART‘the two red flowers’103


(75) viex vie a hni a zzyx nyip buꃧꃧ ꀊꀊ ꀊꀊ ꑍ ꀮflower red those two CL‘those two red flowers’3.4.2.1 Definite nouns with definite article suWhen the DP is definite, D can have the phonetic realization <strong>of</strong> su. CL willadjoin to this definite article via F. <strong>The</strong>refore definite nominals are marked by[CL+F+su]. After CL has incorporated into D, the remnant CLP moves to [Spec, DP],resulting in the desired definite nominal string [NP+AP+NumP+CL+su]. <strong>The</strong>derivation <strong>of</strong> (74) is as follows:(76)suARTviexvie bu a hni/nyipflower CL red/two104


3.4.2.2 Definite nouns with demonstrativesDefinite nominals may also have such word order as N>A>Dem>Num>CL, as isillustrated in (75). In this case, D has no phonetic realization. CL adjoins to F belowthis empty D. <strong>The</strong> co-indexation <strong>of</strong> DemP and NP at [Spec,CLP] renders the NPdefinite, and in turn makes the whole CLP definite. <strong>The</strong> definite CLP moves to[Spec,DP] and checks the strong D feature <strong>of</strong> the null D. <strong>The</strong>refore, we conclude thatthe definite nominal with a demonstrative is marked by [CL+F]. After CL hasincorporated into F, the remnant CLP moves to [Spec, DP].(77)105


<strong>The</strong> relative word order <strong>of</strong> AP/DemP/NumP is derived as follows:(78)<strong>The</strong>se two options for encoding definiteness in Yi is in the same spirit <strong>of</strong> theDoubly Filled Comp Filter (Keyser 1975), which says that C must be null inembedded clauses if [Spec, CP] is not null. <strong>The</strong> Yi data show that D must be null if[Spec, DP] carries the feature [+D]. This explains why demonstratives cannotco-occur with the definite article in Yi.To sum up, we can see that data from Yi show that the CCH can easily capturethe syntactic derivation <strong>of</strong> both definite and indefinite Yi noun phrases, and can easilyexplain the definiteness encoding in the Yi language. All the previous analysesavailable cannot readily capture the syntactic distribution and definitenessinterpretation <strong>of</strong> the Yi data.106


3.4.3 Extended projection <strong>of</strong> indefinite nouns in Mandarin<strong>The</strong> CCH can be readily extended to SVO languages like Mandarin andCantonese. Mandarin indefinite nominals mainly involve numerals and adjectives.(79) san zhang yuan zhuo zi三 张 圆 桌 子three CLround table‘three round tables’We still assume that there is CL-to-F movement, causing CL to play a role indetermining the referential property <strong>of</strong> the extended projection <strong>of</strong> nouns. For SVOlanguages, it is the complement <strong>of</strong> CLP, rather than CLP itself, that moves to [Spec,DP] via [Spec, FP]. 16 In (79) it is the NumP that moves. <strong>The</strong> adjective ‘round’ is notbase-generated along with numerals. Rather they are base-generated with the nounzhuozi (‘table’) so as to form the predication relation between it. <strong>The</strong> [Spec, CLP] isan FP yuanzhuozi (‘round table’), which is derived by predicate inversion. <strong>The</strong> twoderivational steps are shown in the following two tree diagrams.16 We still assume that predicate inversion move takes place here. NumP will undergo predicateinversion A-movement to [Spec, FP] first, and then it undergoes A’-movement to [Spec, DP].107


(80)zhuozitableyuanround3.4.4 Extended projection <strong>of</strong> definite nominals in Mandarin<strong>The</strong> extended projection <strong>of</strong> definite nominals in Mandarin assumes the wordorder <strong>of</strong> Dem>Num>CL>A>N, as is exemplified in (81):108


(81) na san zhang yuan zhuo zi那 三 张 圆 桌 子that three CL round table‘three round tables over there’<strong>The</strong> derivation process is shown in the following tree diagram:(82)round tableyuan zhuozizhangnathatsanthreeCL109


3.5 SummaryIn this chapter, we have examined the syntactic distribution and semanticinterpretation <strong>of</strong> Mandarin, Cantonese and Yi classifier phrases, and review threeanalyses <strong>of</strong> Chinese classifier phrases. Adopting CCH sketched in the previouschapter, we propose an alternative analysis for classifier phrases. <strong>The</strong> analysisproposed here is well supported by empirical data regarding the derivation <strong>of</strong> definiteand indefinite nominal phrases in both Yi and Mandarin.110


CHAPTER 4Possessives in Yi and Mandarin<strong>The</strong> semantics <strong>of</strong> possession can be expressed by a variety <strong>of</strong> syntactic structures. Iwill start this chapter with a background <strong>of</strong> possessive syntax, and end this chapter withan illustration <strong>of</strong> how Yi and Mandarin present us with a unique perspective onpossessive syntax.4.1 Issues <strong>of</strong> possessive syntax<strong>The</strong> possession relation can be realized either lexically or syntactically. In Englishwe have verbs encoding the possession semantics such as possess and own. However, thepossession relation can also be encoded in nominal structures without these verbs. In bothnominal possessives and clausal possessives, syntactic means are available to generatethe possession semantics.4.1.1 Lexical possessionSome verbs can encode possession relation in their argument structure. <strong>The</strong>se verbsare normal two-place predicates, assigning “possessor” and “possessee” theta roles totheir external and internal arguments, shown in (1) and (2).(1) John possesses a big garden.(2) John owns a big garden.111


4.1.2 Syntactic possessionPredicates in the possessive sentences may not be possession verbs like possess orown. Rather they may be non-lexical verbs. For example, in English we can use thecopular verb be:(3) <strong>The</strong>se two books are <strong>of</strong> John’s.This pattern is also observed in other languages such as Tagolog:(4) Mayrelos ang naanai [Freeze 1992: 577]COP-watch ART mom‘Mom has a watch.’It is difficult for us to get the possession reading directly from the lexical meaning <strong>of</strong>the copular verb. <strong>The</strong>refore, it is not unreasonable to assume that the possessiverelationship derives from the predication relation within a small clause, which means thatthe possession reading derives from the syntactic structure <strong>of</strong> the whole sentence ratherthan the lexical meaning <strong>of</strong> the matrix verb. Besides the copular verb, the verb have canalso be used in English possessive sentences:(5) John has three sisters.Although we can argue that have is a lexical verb just like own and possess, variousanalyses (Freeze (1992); Kayne (1994); den Dikken (1997)) have been proposed tochallenge the assumption that have is a lexical verb. All these analyses share theassumption that the predication relation which determines the possession semantics is notestablished by a verbal head. Under these analyses the verb have which surfaces inpossessive-have sentences like (5) is the spell-out <strong>of</strong> a copular verb (like be in English),112


which is a raising verb, taking a small clause as its complement. Within the small clause,the predication relation is established.If we view possession as a predication relation within a small clause, there are twoways to encode the predication relation. <strong>The</strong> first way is to view the possessor as thesubject <strong>of</strong> the predication relation, whereas the possessee is the predicate. <strong>The</strong> secondway is to view the possessor as part <strong>of</strong> the predicate, and possessee as the subject. <strong>The</strong>first way encodes the predication relation in the canonical subject-predicate word order;therefore, we call it predication approach. <strong>The</strong> second way treats the possessor as thecomplement <strong>of</strong> a preposition, indicating the location <strong>of</strong> the possessee; therefore we call itlocative/existential approach.4.1.2.1 Predication approachUnder the predication approach, the possessor is the subject <strong>of</strong> the predicationrelation, whereas the possessee is the predicate. This view gains support from Hungarianpossessives, in which the possessee shares the same person and number features with thepossessor subject, just in the same manner that common predicates agree with subjects <strong>of</strong>finite clauses in person and number features .(6) a. as en hiz-omthe 1sg-NOM house – 1sg‘my house’113


. En isz-om1sg-NOMdrink – 1sgIn the Hungarian possessive (6a), we find a linear sequence <strong>of</strong> a definite determiner (as),a nominative possessor (en) and a possessed noun (hiz) with an agreement suffix (-om).<strong>The</strong> agreement morpheme indicates that there exits an agreement relationship betweenthe possessor and the possessee in person and number, an agreement relationship thatessentially exists between a nominative subject and a finite verb in a clause such as (6b).This phenomenon can also be observed in Altaic languages such as Uyghur. 1(7) [[Tursun-ning uka-si]-ning] muallim-i (Litip 2004: 284)Tursun-GEN brother-3sg-GEN teacher-3sg‘Tursun’s brother’s teacher’ (‘the teacher <strong>of</strong> Tursun’s brother’)Tursun, as a proper name, is featured by [3 rd person, singular]. –si, suffixed to thepossessee, is the inflectional morpheme with the feature [3 rd person, singular], showingthe person and number agreement with the possessor. <strong>The</strong> agreement relation can beschematized as follows. –si agrees with Tursun, and –i agrees with uka. 2(8) [[Tursun-ning uka-si]-ning] muallim-i1 In Uyghur, ‘-i’ and ‘-si’ are allomorph <strong>of</strong> the inflectional morpheme indicating Person and Number feature[3 rd person, singular] <strong>of</strong> the Possessor. ‘-si’ is used when the possessee noun ends up with vowels, and‘-i’ is used when the possess noun ends up with consonants.2 <strong>The</strong> arrows show the agreement relation between the elements at the head and the tail <strong>of</strong> the arrow.114


<strong>The</strong>refore, we can claim that Hungarian and Uyghur possessive nominals have thefollowing structure:(9) [ DP (Spec) [ D’ D [ AgrP Possessor [ Agr’ Agr Possessee]]]]<strong>The</strong> possessee N enters Numeration with suffixes indicating person and numberfeatures. In the derivational process, the possessee NP checks the person and numberfeatures <strong>of</strong> Agr. Either the possessee N undergoes feature-Move and check, thus no overtN-to-Agr movement, or the features <strong>of</strong> N pied-pipe all the other features <strong>of</strong> N to Agr, thusN will adjoin to Agr. <strong>The</strong> possessor subject at [Spec, Agr] determines the person andnumber features <strong>of</strong> Agr. This Agr head can be argued to be a kind <strong>of</strong> nominal INFL,responsible for forming the possession-relevant predication relation. We will incorporatethis insight into our analysis <strong>of</strong> nominal structure. However, we will simply dub it as a115


functional projection (FP), leaving the INFL-feature <strong>of</strong> the functional head in general. 3Besides using the nominative possessor, Hungarian can also use dative possessor, as isshown in (10c).(10) a. a Mari kalap-ja-ithe Mary-NOM hat-POSS.3SG-PL‘Mary’s hats’b. Mari-nak a kalap-ja-iMary-DAT thehat-POSS.3SG-PL‘Mary’s hats’c. Mari-nak van-nak kalap-ja-iMary-DAT be-3PLhat-POSS.3SG-PL‘Mary has hats.’ (Szabolcsi 1994: 223)Szabolcsi (1994) claims that the dative possessor Mari-nak is base-generated withinthe nominal phrase, and undergoes possessor extraction to the position left to the verbthrough [Spec, DP], as is shown in (11):3 In Hungarian and Uyghur, the functional head apparently is featured by phi-Agreement. However, inMandarin and other classifier languages, there is no apparent morphological evidence for the existence <strong>of</strong>such a Agr projection, although we assume that there still exists a functional projection between DP andCLP. <strong>The</strong> function <strong>of</strong> this functional projection is, in essence, nominal INFL, which is responsible forencoding the temporal-spacial properties <strong>of</strong> the whole nominal phrase. In particular, this nominal INFLelement is shown to be essential for the specificity-encoding in classifier languages discussed in this thesis.116


(11) Mari-nak i van-nak [ DP t i [D [ (N+I)P t i kalap-ja-i]]]Mari-DAT be-3PL hat-POSS.3SG-PL‘Mary has hats.’ (Szabolcsi 1994: 180)In this case, D takes a (N+I)P as its complement. <strong>The</strong> possessor is base-generated at[Spec, (N+I)P], and the possessee is the complement <strong>of</strong> (N+I)P. According to thisapproach, sentence (12) will have the structure as in (13):(12) John has a house.(13) BE [ DP D [ (N+I)P John a house]This possessive construction is reminiscent <strong>of</strong> the predication relation holding within theFrench double object construction as in (14) and the English double object constructionas in (15).(14) la [ CP voiture j [de [ IP Jean [I [e] j … (Kayne (1994: 102)the car <strong>of</strong> Jean‘Jean’s car/the car <strong>of</strong> Jean’s’(15) Mary gave [ SC [John] a house](14) and (15) share the same structure <strong>of</strong> a small clause involving the possessor andthe possessee. In (14) Jean and voiture forms a non-finite IP — a small clause. In (15)John and a house forms a small clause. <strong>The</strong> idea that double object construction (DOC)involves a possessive small clause is also shared in Larson (1988) and Bowers (1993),shown in the following section.117


4.1.2.1.1 Larson’s analysis <strong>of</strong> DOCLarson’s (1988) analysis is that DOC is derived from dative construction by anoperation called Dative Shift. A simple dative like (16a) derives from an underlyingstructure in which the verb and its indirect object form a constituent. It involves aclause-like VP whose “subject” is the direct object a letter and whose “object” is theindirect object (to) Mary. Verb is raised up.(16) a. John sent a letter to Mary.b. John sent i [ VP a letter [ V’ t i to Mary]]c. John sent [Mary a letter].Larson argues that double object construction is derived from the dative construction by akind <strong>of</strong> VP-internal passivization operation called Dative Shift. <strong>The</strong> former indirectobject Mary becomes a derived VP “subject” and the former direct object a letter isdemoted from the specifier position to an adjunct position. <strong>The</strong>refore, in the DOC (16c),[Mary a letter] can be understood as a passivized small clause, with Mary being theraised subject.4.1.2.1.2 Bowers’ analysis <strong>of</strong> DOCBowers (1993) argues that the double object construction is in fact a doublepredication structure. For example:(17) John gave Mary the book.118


Under this analysis, the verb give contains a [+cause] feature. Basically, we canargue that the two semantic primitives <strong>of</strong> the verb give is [+cause] and HAVE. <strong>The</strong>derivation <strong>of</strong> DOC is based on these two primitives. <strong>The</strong> verb give first gets raised to thelower Pr position. It then adjoins to the higher V position so as to check themorphological feature [+cause] with its own [+cause] feature. <strong>The</strong> subject <strong>of</strong> the lowerPrP then raises to the object position in the higher PrP to get Case. 4 Finally, the verb givemoves up to the Pr position in the upper PrP.According to this view, DOC is virtually identical in underlying form to causative4 Bowers (1993) claims that [Spec, PrP] is not a Case-marked position. [Spec, IP] is the nominativeCase-marked position, and [Spec, VP] is the accusative Case-marked position.119


constructions with make. <strong>The</strong> only difference between the two constructions is that theexplicit causative has a phonological specified causative verb in the upper clause.(18) [ PrP John made the lions i [ PrP t i eat j [ VP the meat t j ]]].Both (17) and (18) contain a possessive small clause—the lower PrP. Bowers thusexplicitly claims that under the covert causative verb there is a proposition [Maryowns/has the book]. Both Larson’s and Bowers’ analyses <strong>of</strong> DOC involve a SC, whichinstantiates the possession semantics.4.1.2.2 Existential approach<strong>The</strong> other approach to possessive syntax is to view possessive relation as essentiallylocational. This approach also hinges upon small clause structure, but the subject <strong>of</strong> thepredication in this case is the possessee and the predicate is the possessor.Den Dikken (1998) assumes that the semantic relation holding between thepossessor and the possessed object in possessive-have sentences is established within asmall-clause construction in which the possessed object constitutes the subject <strong>of</strong>predication and the possessor is the complement <strong>of</strong> a PP which constitutes the predicate.<strong>The</strong> semantic relation and the syntactic derivation process can be illustrated by thefollowing example:120


(19) a. John has three sisters.b. [ IP Spec [ I’ I [ XP [ DP three sisters] [ X’ X [ PP P[ DP John]]]]]]c. [ IP [ PP t j [ DP John]] i [ I’ [P j +X k +I (=have)] [ XP [ DP three sisters] [ X t k t i ]]]]<strong>The</strong> derivation <strong>of</strong> the surface structure (19c) involves movement <strong>of</strong> the predicate PP to[Spec,IP] and incorporation <strong>of</strong> the head X (Pr in Bowers’ predication syntax) <strong>of</strong> the smallclause and the preposition P into I, forming a complex head which is spelled out as have.Raising <strong>of</strong> the predicate PP to [Spec, IP] is treated by den Dikken as an instance <strong>of</strong>predicate inversion. <strong>The</strong> above derivational process can be illustrated in (20):(20)In den Dikken (1998), the structure <strong>of</strong> a genitive construction like (21a) involves asmall-clause established between an NP subject and a PP predicate, shown in (21b).121


(21) a. John’s dogb. [ DP Spec [ D’ D [ FP Spec [ F’ F [ XP [ NP dog] [ X’ X [ PP P [ DP John]]]]]]]]c. [ DP Spec [ D’ D[ FP [ PP t k [ DP John]] i [ F’ [P k +X j +F (=’s)] [ XP [ NP dog] [ X’ t j t i ]]]]]]In a way parallel to the sentential case shown in (19), the surface structure <strong>of</strong> (21a) isderived by DP-internal predicate inversion. <strong>The</strong> PP predicate lands in the specifier <strong>of</strong> thefunctional head F, which gets spelled out as -’s as a consequence <strong>of</strong> the incorporation <strong>of</strong>the head <strong>of</strong> the small clause X and <strong>of</strong> the preposition P heading the predicate PP.According to this approach, HAVE is considered to be a complex <strong>of</strong> an abstractcopula BE and an adpositional element.(22) a. BE [ PP [ a house] P John]b. BE+Pi [ PP [a house] ti John]After the incorporation <strong>of</strong> P to BE, predicate inversion will take place, resulting in thefollowing surface structure:(23) John has a house.This approach captures the fact that possessive sentences and existential constructionscan be cast in the same mold. Mandarin has the following minimal pair.(24) zhuo shang you yi ben shu.桌 上 有 一 本 书 。desk top have one CL book‘<strong>The</strong>re is a book on the table.’122


(25) Zhangsan you yi ben shu.张 三 有 一 本 书 。Zhangsan have one CLbook‘Zhangsan has a book.’<strong>The</strong> same pattern is also observed in Yi. 5(26) ngop jiet bbap ga go ke Mat Hxie hmi xip ma jjo.ꉪ ꏤ ꁡ ꇤ ꇬ ꈌ ꂵ ꉌ ꂓ ꑠ ꂷ ꐥ.our home village at dog Mathxie name such CL have‘<strong>The</strong>re was such a dog named Mathxie in our home village.’(27) ax yi cyx ma i dix nyip ggu jjo.ꀉꀉ ꋋ ꂷ ꀂ ꄁ ꑍ ꈬ ꐥchild this CL coat two CL have‘This child has two coats.’4.1.2.3 Possessor Raising approachThis approach views sentential possessives as a derivation <strong>of</strong> DP possessives.Szabolcsi (1994) argues that in Hungarian possessive sentences with BE, the possessor isbase-generated within the nominal projection <strong>of</strong> possessee, and later undergoes5 Example (26) is from Walters and Atqi (2006: 6).123


possessor-raising. This approach is motivated by the observation that the possessor agreeswith the possessee in person and number just as the subject agrees with the verbs:(28) a. En isz-om1sg-NOM drink-1sg‘I drink.’b. as en kalap-omthe 1sg-NOM at-1sg‘my hat’Besides using the nominative possessor as in (28b), Hungarian can also use dativepossessor as in (10), repeated here as (29):(29) a. a Mari kalap-ja-ithe Mary-NOM hat-POSS.3SG-PL‘Mary’s hats’b. Mari-nak a kalap-ja-iMary-DAT thehat-POSS.3SG-PL‘Mary’s hats’c. Mari-nak van-nak kalap-ja-iMary-DAT be-3PLhat-POSS.3SG-PL‘Mary has hats.’ (Szabolcsi 1994: 223)124


Szabolcsi (1994) claims that Hungarian possessive sentences are derived frompossessive DPs: the dative possessor Mari-nak is base-generated within the nominalphrase, and undergoes possessor extraction to the position left to the verb through [Spec,DP], as is shown in the following:(30) Mari-nak i van-nak [ DP t i [D [ (N+I)P t i kalap-ja-i]]]Mari-DAT be-3PLhat-POSS.3SG-PL‘Mary has hats.’ (Szabolcsi 1994: 180)Possessors can occupy two different prenominal positions. In (29b), the possessor isin the dative, rather than in the nominative, and precedes, rather than follows, the definitearticle. When the possessor follows the determiner, there is no case marker, whichcorresponds to nominative in Hungarian. <strong>The</strong> pre-determiner possessor is dative-marked.Szabolcsi argues that the possessor moves from the post-determiner position to thepre-determiner position. Based on the morphological marking on the possessee, Szabolcsi(1994) claims that [+poss, agr] inflection on the [possessee] noun licenses nominativecase for the possessor, just like [+tense, agr] inflection on the verb licenses nominativecase for the subject. 6 <strong>The</strong> structure for nominal possessives then should be as follows:6 Pushing the analogy further, Szabolcsi argues that we may assimilate plain noun phrases to [-tense, agr]infinitives, and assume that they have [-poss, agr] inflection, rather than no inflection.125


(31)Here the head <strong>of</strong> the projection is the inflected noun N+I as a whole. Recall that theinflectional element may be [-poss, agr] or [+poss, agr]. [+poss] changes the argumentstructure <strong>of</strong> the [-poss] noun in the manner derivational affixes do:(32) dog[-poss] : dog[+poss] = victim : victimizeThis is another case <strong>of</strong> predicativization as discussed in Chapter 2. That is, the [+poss]feature creates a subject and establishes the possession relation between the subject andthe possessee. Szabolcsi also argues that [Spec, DP] is an operator position, just like[Spec, CP], and this position is the place to license the dative Case <strong>of</strong> possessor. <strong>The</strong>possessor must get raised to [Spec, DP] before it moves out <strong>of</strong> the DP. When thepossessor is in the DP, either in nominative Case or dative Case, the DP is specific. For apossessive DP to be non-specific, the possessor has to move out <strong>of</strong> DP.Since existential constructions show specificity effect (Enç 1991), which says thatexistential constructions require non-specific indefinite arguments. <strong>The</strong> similarity inreferentiality condition imposed on the possessor-extracted possessive constructions andthe referentiality condition imposed on existential clauses leads Szabolcsi to concludethat the have-sentence in Hungarian is existential sentence.126


As for the encoding <strong>of</strong> definiteness <strong>of</strong> Hungarian nominals, Szabolcsi proposes thefollowing interpretation principles based on the tree diagram in (31):a. DetP determines both the quantification and the definiteness <strong>of</strong> the noun phrasethrough determining these properties <strong>of</strong> (N+I)P.b. DetP may be phonetically empty.c. a(z) 'the' or Ø 'a, some' is selected for D in agreement with the definiteness <strong>of</strong>D's complement (N+I)P.This means that the encoding <strong>of</strong> definiteness is not solely by D. Instead it is decidedjointly by D and its complement. Kayne (1994: 95) makes similar remarks to thefiniteness <strong>of</strong> clauses. Observing that relatives with fully finite verbs are common inpostnominal relatives, Kayne predicts that full finiteness is normally incompatible with IPbeing split <strong>of</strong>f from C, because finiteness required incorporation <strong>of</strong> I to C in the overtsyntax and that relation could not be reconstructed after overt movement <strong>of</strong> IP away fromC. Finiteness requires incorporation <strong>of</strong> I to C. This is very important. We would like topropose that similarly definiteness requires incorporation <strong>of</strong> Poss (N+I) to D. <strong>The</strong>incorporation is licensed by the agreement between D and the head <strong>of</strong> its complement127


(N+I).Szabolcsi further argues that DetP may consist <strong>of</strong> features [+ definite] and [+specific], and these determine the choice <strong>of</strong> a(z) or Ø in D. [+def] and [-def,+spec](N+I)P's select a(z) 'the,' and [-spec] (N+I)P's select Ø. Compare (33) with (34):(33) Mari-nak van-nak kalap-ja -i.Mari-DAT be -3PLhat -POSS.3SG-PL(-NOM)'Mari has hats.'(34) Van-nak kalap-ok.be -3PL hat -PL(-NOM)'<strong>The</strong>re are hats.'Sentence (34) is the standard existential sentence in Hungarian, and it contains thesame verb as the possessive sentence <strong>of</strong> (33). Syntactically speaking, the only differencebetween the two is that (34) has a [-poss] nominal INFL, and (33) a [+poss] one, with itspossessor extracted to the subject position <strong>of</strong> the matrix clause. We know that existentialverbs cross-linguistically require a non-specific indefinite argument; this is what weknow as definiteness effect. Sentences (33) and (34) lead Szabolcsi to the followinggeneralization:<strong>The</strong> "HAVE sentence" in Hungarian is an existential sentence with a [+poss]nominative argument. Given that (i) the existential verb requires a non- specificindefinite argument and (ii) a [+poss] DP has a non-specific indefiniteinterpretation only if its possessor is extracted, possessor extraction in the HAVE-128


sentence is obligatory. Have-possessives are existential sentences with possessorextraction.Szabolcsi (1994: 550)Szabolcsi argues that a possessive-have sentence is derived by the possessorextraction from a DP, as shown in (35) and exemplified in (36).(35) Deriving possessive sentences from nominal possessives:Possessor-DAT BE (existential) [ DP t Possessor-DAK Possessee](36) Mari-nak van-nak kalap-ja -i.Mari-DAT be -3PL hat -POSS.3SG-PL(-NOM)‘Mari has hats’ (<strong>The</strong>re exist hats that stand in some relation R to Mari.)Kayne (1994) also argues that possessive sentences involve a copular head with aD/PP complement, and that possessive HAVE is not a primitive lexical item, but rather isderived from an abstract copular verb BE and an abstract ‘prepositional’ determiner (D/P).<strong>The</strong> possessor moves through [Spec, DP] to the surface subject position. D/P incorporatesinto BE, and BE plus the incorporated D/P surfaces morphologically as HAVE.(37) a. John has a sister.b. John i BE [ DP D [ AgrP t i [ a sister]]]c. John i BE+D j [ DP t j [ AgrP [ t i a sister]]]129


<strong>The</strong> derivational process can be illustrated by the following tree:(38)4.1.2.4 Interium Summary: a unified approach<strong>The</strong> three approaches reviewed above share the same idea that they all assume asmall clause representing the predication relation between the possessor and the possess.For the predication approach, D takes a (N+I)P (a small clause, or PrP in Bowers’ terms)as its complement. <strong>The</strong> possessor is base-generated at [Spec, (N+I)P], and the possesseeis the complement <strong>of</strong> (N+I)P. According to this approach, (39a) has the structure (39b):(39) a. John has a house.b. BE [ DP D [ (N+I)P John a house]This approach explains the feature agreement between the possessor and the possessee,but it fails to capture the similarity between possessor-possessives (possessives withanimate possessors) and existential possessives. <strong>The</strong>refore, this thesis will adopt the130


existential approach to possessives, since it captures the empirical observation in both Yiand Mandarin that possessive sentences and existential constructions are inherentlyrelated to each other. We propose that Possessor is not base-generated at [Spec,PrP]. It isbase-generated as the complement <strong>of</strong> an abstract PP, which is the complement <strong>of</strong> PrP.<strong>The</strong> possessor or the locational phrase can undergo predicate inversion to reach [Spec,PrP]. It can further undergo A’-movement to [Spec, DP], which serves as the only escapehatch for the possessor phrase to get raised outside DP. Viewed in this way, the threeapproaches are in fact in the same direction.4.2 Nominal and Clausal possessives in Yi<strong>The</strong> unified approach to possessives sketched above can find empirical support fromboth Yi and Mandarin possessives. In this section we will examine nominal and clausalpossessives in Yi.4.2.1 Nominal possessives in Yiforms:<strong>The</strong> semantic relation <strong>of</strong> possession in Yi can be expressed in different nominal(i)Possessor NP1+ Possessee NP2(40) mu gat i dixꃅꃅMugaꀂ ꄁcoat‘Muga’s coats’131


In this case, two nominals form the possessive construction with the first nominal servingas the possessor and the second as the possessee. <strong>The</strong>re is no linking morphemeindicating the possession relation.(41)i dixcoatmu gatMugaWe assume there is an abstract P, which complement-selects the possessor. This abstractP first incorporates into CL. 7 <strong>The</strong> complex [P+CL] head further incorporates into F,extending the domain and enabling the predicate inversion. F is featured by [+POSS].After the predicate inversion Move, the possessor finally reaches [Spec, FP].(ii)Possessor NP1+ Possessee NP2 +AP/Num CL-øThis form is different from (i) in that it is more complex, involving the use <strong>of</strong> cardinalnumerals, adjectives and classifiers.7 It is more appropriate to say that the head is Pr rather than CL. Pr can be phonetically realized as CL.However, when Pr does not complement-selects NumP or AP, Pr may be null phonetically. That is to say, Prhas two realizations at PF: the overt realization <strong>of</strong> CL and the covert realization as a zero morpheme.132


(42) mu gat i dix nyip gguꃅꃅ ꀂ ꄁ ꑍ ꈬMuga coat two CL‘Muga’s two coats’(43) mu gat i dix a shyt gguꃅ ꃅ ꀂ ꄁ ꀊ ꏀ ꈬMuga coat new CL‘Muga’s new coats’We would regard this possessive construction as involving two predication relations.i dix [ nyip/ a shyt] ggu ( ‘coat two/new CL’) realizes the first predication relation, asis shown in (44):(44)i dix ggu nyip / a shytcoat CL two / new133


<strong>The</strong>n the resultant DP forms the second predication relation with the possessor, as isshown in (45). In the first predication relation, Pr is phonetically realized as CL; whereasin the second predication relation, Pr is phonetically null.(45)i dix nyip ggucoat two CLmu gatMuga<strong>The</strong> possessee (Subject <strong>of</strong> the upper PrP) in this case is a DP. This DP (44)contains a CLP (PrP), the head <strong>of</strong> which (CL) incorporates into F, which is [-POSS],indicating that the noun at [Spec,CLP] is either non-specific or specific. 8 After CL-to-Fincorporation, the whole remnant CLP will undergo A’-movement to [Spec, DP]. It is notpossible for the CLP to undergo A-movement to [Spec, FP]. <strong>The</strong> remnant PrP is acomplete proposition rather than a predicate. <strong>The</strong>refore, it cannot undergo predicateinversion move. Yi in this respect differs significantly from Mandarin and Jingpo. <strong>The</strong>A’-movement <strong>of</strong> the whole CLP to [Spec, DP] may be due to the strong topic-prominent8 Only when the noun at [Spec, CLP] is specific can the F form a concordance relation with D to determinethe definiteness <strong>of</strong> the whole nominal.134


property <strong>of</strong> Yi.(iii)Possessor NP1+Possessee NP2 +AP/Num CL-suThis form differs from (ii) in that it contains the definite article su. Hence the possessee isdefinite in this construction. For example, (46) refers to a unique coat possessed byMuga.(46) mu gat i dix a shyt ggu suꃅꃅ ꀂ ꄁ ꀊ ꏀ ꈬ ꌠMuga coat new CL ART‘the new coat <strong>of</strong> Muga’s’Under our analysis, there is CL-to-F-to-D incorporation in this case. <strong>The</strong> NP at[Spec, CLP] is specific. F is featured by [-POSS] (to be more specific, it is [+specific,-definite]). Since F is [+specific], it can form a concordance relation with D. <strong>The</strong> wholeCL-F-D chain determines the referential property <strong>of</strong> the NP at [Spec, CLP].(iv)Possessor NP1+Possessee NP2 +AP suIn this form, there is no classifier. That is to say, both <strong>of</strong> the two predication relationsinvolved in the possessive phrase lack the overt realization <strong>of</strong> the head Pr(47) mu gat i dix a shyt suꃅꃅ ꀂ ꄁ ꀊ ꏀ ꌠMuga coat new ART‘Muga’s new coat(s)’135


In this case there is no indication <strong>of</strong> the number feature <strong>of</strong> the possessee within thewhole phrase. We can claim that the default form <strong>of</strong> Pr is a phonetically null morpheme.<strong>The</strong> overt realization <strong>of</strong> Pr as CL is driven by the requirement to specify the numberfeature <strong>of</strong> the possessee.(48)i dixcoata shytnewTo sum up, the nominal possessives in Yi involves two layers <strong>of</strong> predicationrelations. First, if the possessee phrase contains a CL or a D, we assume that there is aninternal predication relation within this possessee DP. This possessee DP then enters thepredication relation with the possessor, which reaches [Spec, FP] via the predicateinversion Move.4.2.2 Clausal possessives in YiYi has lexical verbs to indicate possession. <strong>The</strong> most widely used verb is bbop (ꁨ),which is the counterpart <strong>of</strong> English possess or own.136


(49) ix yi ne lap bbu ji bbop.ꀁꀉ ꆏ ꇂꇂ ꏢ ꁨ.brother TOP ox CL have‘<strong>The</strong> brother has an ox.’Some existential verbs can also be used to indicate possession. According to Walters& Ndaxit (2006), there are fourteen existential verbs in Liangshan Yi. Among all theexistential verbs, jjo (ꐥ) is a versatile verb to express location, possession, and existence.(i)Location(50) vut nyop gox jjo jjo? 9ꃴꑘ ꇫ ꐥ ꐥVutnyop here be be‘Is Vutnyop here?’(51) cyp ax mo ix go jjo.ꋍ ꀉꂿ ꀁ ꇬ ꐥ .his mother home at be‘His mother is at home.’9 <strong>The</strong> reduplication <strong>of</strong> the main verb in a sentence is a method to form yes-or-no questions in Yi.137


(ii)Possession(52) ax yi cyx ma i dix nyip ggu jjo.ꀉꀉ ꋋ ꂷ ꀂ ꄁ ꑍ ꈬ ꐥchild this CL coat two CL have‘This child has two coats.’(iii)Existence(53) bbap ga go kex ma jjo.ꁡ ꇤ ꇬ ꈋ ꂷ ꐥ .village at dog CL have‘<strong>The</strong>re is a dog in the village.’<strong>The</strong> difference between location and existence is apparent. <strong>The</strong> noun phrase in (i)must be a definite noun phrase, and the verb jjo is best translated as copular verb BE. <strong>The</strong>noun phrase in (iii) must be a non-specific noun phrase, since there is a definiteness effectconstraint for existential sentences. We would call sentence like (53) existentialpossessive (E-possessive), and argue that both possessor-possessives and existentialpossessives are derived from nominal possessives with the possessor extraction.We will show that the existential possessive in Yi is essentially parallel to theHungarian clausal possessive discussed in Szabolcsi (1994). Following Szabolcsi, wepropose that the E-possessives have a nominal base structure and the possessor undergoesraising from its base-generated position inside the nominal domain to the subject <strong>of</strong> thewhole sentence.138


(54) Mu gat i dix nyip ggu jjo.ꃅ ꃅ ꀂ ꄁ ꑍ ꈬ ꐥMugat coat two CL have‘Mugat has two coats.’(55) nga hnip mop cyp ma jjo.ꉢ ꅫꃀ ꋍ ꂷ ꐥ .I sister one CL have‘I have one sister.’I will claim that these sentences are derived from the possessive DPs. <strong>The</strong> possessorwill be extracted from [Spec, DP] to [Spec, IP], and the D will incorporate into theexistential copular verb BE surfacing as jjo (‘have’). <strong>The</strong> derivation <strong>of</strong> (54) can be shownin (56):(56) [mu gat] i [ DP t i [t j ] t i i dix nyip ggu ] [D j +BE = jjo]ꃅ ꃅ ꀂ ꄁ ꑍ ꈬ ꐥMugat coat new CL have‘Mugat has two coats.’<strong>The</strong> derivational process involves two steps. First, the possessor will undergopredicate-inversion to [Spec, FP], shown in (57).139


(57)i dix nyip ggucoat two CLmu gatMugaAfter that, the possessor at [Spec, FP] undergoes A’-movement to [Spec, DP]. <strong>The</strong>abstract copular verb is a raising verb, taking a DP/CP as the complement. D adjoins tothis copular verb, surfacing as jjo. <strong>The</strong> FP moves up to [Spec, CP] via [Spec, DP],resulting in the Yi clausal possessives like (54) and (55). This derivational process isillustrated by (58):140


(58)<strong>The</strong> difference between Hungarian and Yi possessives is that in Hungarian it is thepossessor that is raised to [Spec, DP], whereas in Yi it is the FP containing both thepossessor and possessee that is raised to [Spec, DP]. <strong>The</strong> derivation is as follows:(59)141


4.3 Mandarin PossessivesIn this section we will examine the syntactic distribution <strong>of</strong> Mandarin possessivesand the encoding <strong>of</strong> definiteness <strong>of</strong> possessives. Mandarin nominal possessives involvethe use <strong>of</strong> a multi-functional morpheme de ( 的 ), and Mandarin clausal possessivesinvolve the use <strong>of</strong> the verb you ( 有 ). <strong>The</strong> Mandarin nominal and clausal possessives areexemplified in (60) and (61).(60) Xiaoming de yi ben shu小 明 的 一 本 书Xiaoming DE one CL book‘a book <strong>of</strong> Xiaoming’s’(61) xiaoming you yi ben shu.小 明 有 一 本 书 。Xiaoming have one CLbook‘Xiaoming has a book.’4.3.1 Issues <strong>of</strong> Mandarin possessivesIn Mandarin, we have two alternative ways to express nominal possessives.(62) Possessor>Dem>Num>CL>N(63) Dem>Num>CL>Possessor>N142


(64) zhangsan de na san ben shu张 三 的 那 三 本 书Zhangsan DE that three CL book‘those three books <strong>of</strong> Zhangsan’s’(65) na san ben zhangsan de shu那 三 本 张 三 的 书that three CL Zhangsan DE book‘those three books <strong>of</strong> Zhangsan’s’(64) and (65) are different in that in (64) the possessor c-commands the whole stringDem>Num>CL>N, whereas in (65) the possessor only c-commands N. (66) and (67)illustrate the different derivations <strong>of</strong> (64) and (65).(66) a.shubookbenCLnathatsanthree143


.na san ben shu zhangsanthat three CL book zhangsan(67) a.shubookzhangsanzhangsan144


.benCLZhangsan-de shuZhangsan-GEN booknathatsanthreeFrom the tree diagrams we see that the difference between (62) and (63) is that (62) is aPossP, whereas (63) is a DP.<strong>The</strong> derivation <strong>of</strong> (62) is shown in (66). First <strong>of</strong> all, there is a DP projection 那 三 本书 (‘those three books’). This DP then enters the possession relation with regard to thepossessor DP 张 三 , as is shown in (66a), where the head Pr adjoins to F [+POSS],enabling the possessor DP 张 三 to be raised to [Spec, FP] via predicate inversion move,as is shown in (66b). It is assumed that the complex functional head [Pr+F] surfaces asde.<strong>The</strong> derivation <strong>of</strong> (63) is shown in (67). <strong>The</strong> whole projection is a DP projection. <strong>The</strong>derivation is also in two steps. First, a PossP 张 三 的 书 (‘Zhangsan’s book(s)’) is145


generated as in (67a). Whether this PossP projects further into DP needs further evidence.We simply assume that the PossP further projects into DP and serves as subject <strong>of</strong> thepredication, occupying [Spec, PrP], as is shown in (67b).4.3.2 Remarks on different approaches to Mandarin deMandarin de ( 的 ) is very versatile. Any attempt to solve the internal syntax <strong>of</strong> theChinese DP cannot stay away from this morpheme. <strong>The</strong> significance <strong>of</strong> this morphemecan be best illustrated by its wide distribution in almost all kinds <strong>of</strong> nominal constructionsin Mandarin Chinese. It can occur with possessive constructions, appositive constructions,relative clauses, pre-nominal adjective phrases, and pre-nominal prepositional phrases.<strong>The</strong>re are three major analyses for this morpheme.4.3.2.1 Ning (1996)Ning (1996) observes that there are four types <strong>of</strong> “de-construction” in Chinese asillustrated in (68) – (71).(i)Type A: [ NP XP+de+ ø] (the bare “de-construction”)(68) [ NP [ XP ta xihuan e] de ø]他 喜 欢 的he like DE‘those he likes’146


(ii)Type B: [ NP XP+de+ NP] (the <strong>Complex</strong> NP construction)(69) [ NP [ XP ta xihuan e] de [ NP na ben shu]]他 喜 欢 的 那 本 书he like DE that CL book‘the book that he likes’(iii)Type C: [ NP XP+de+ NP] (Resumptive <strong>Complex</strong> NP construction)(70) [ NP [ XP wo jian guo ta yi ci] de [ NP na ge gu niang]]我 见 过 她 一 次 的 那 个 姑 娘I see EXP her one time DE that CL girl‘the girl that I saw once.’(iv)Type D: [ NP XP+de+ NP] (NP-complement construction)(71) [ NP [ XP tamen yao jiehun ] de [ NP xiaoxi]]他 们 要 结 婚 的 消 息they will marry DE news‘the news that they are going to get married’<strong>The</strong> common property <strong>of</strong> these de-constructions is that they all have the externalsyntactic status <strong>of</strong> an NP. For Type A and Type B, the XP contains a gap as indicated by e,which is uniformly identified as a typical operator-variable relative clause. Type A isanalyzed by Ning as Chinese free relatives.147


Ning reviews two major analyses <strong>of</strong> de in Chinese. <strong>The</strong> first analysis (Huang 1982,Li 1985) treats de as a Case marker. <strong>The</strong> second analysis (Huang 1982, Ning 1993) treatsde as C with [-wh, +Pred] feature. Ning argues that de cannot be treated as a Case marker,because what de pre-selects is a clause, which does not require Case. He resorts topredicate inversion to explain the following case:(72) Zhangsan de shu张 三 的 书zhangsanDE book‘Zhangsan’s book(s)’In this case, Zhangsan de is the clausal predicate, which merges with another de, and one<strong>of</strong> the de’s is deleted due to phonological repetition. In this way he unifies the syntacticstatus <strong>of</strong> de as C.4.3.2.2 Simpson (1997, 2002)Simpson (2002) analyses de as a D-element. Adopting Kayne’s LCA, Simpsonargues that the only possible analysis for de would seem to be that de in fact is adeterminer occupying D, with the IP-RC raised to Spec DP. Different analyses <strong>of</strong> derender derivational processes <strong>of</strong> relative clauses different. One is adjunction process, as isshown in (73); the other is head-raising process, as is shown in (74).148


(73)wo mai de shuI buy DE book(74)deDEwo mai shuI buy book149


Simpson argues for the raising analysis that the whole structure follows the D-CPhypothesis, as is proposed by Kayne (1994), 10 and then the NP 书 (book) is moved to[Spec,CP], leaving a trace in object position. After that, the whole TP is raised to[Spec,DP]; thus (75c).(75) a. D CPb. D [ NP i C TP t i ]c. TP j D [ NP C t j ]What is called the relative clause, according to Kayne, is the IP that has been raised to[Spec,DP], stranding a zero complementizer. Simpson (1997) argues de, as a determiner,possesses the common property <strong>of</strong> highly selective enclitics. That is, the Chinese de onlyattracts a clausal IP. This provides a strong motivation for the IP movement in MandarinChinese: the movement <strong>of</strong> the IP to [Spec,DP]. It can be said to be triggered by a need toprovide phonological support for de to satisfy its enclitic feature. Simpson also suggeststhat the enclitic feature <strong>of</strong> de can also be explained in Minimalist model: de is assumed tobe base-generated with strong v-(or T-) feature which can be satisfied when a clauseheaded by an element with v-feature is attracted to the specifier <strong>of</strong> DP.Simpson argues that although de is D, it does not encode definiteness. 11 Some othermeans <strong>of</strong> encoding have to be used. Simpson argues that definiteness is specified at alower position, which hosts demonstratives and certain quantifiers.10 After examining N-initial and N-final relative clauses, Kayne (1994: 94) assumes that UG makesavailable for relativization a D CP structure, where CP is the complement <strong>of</strong> D.11 In order words, we can say that the function <strong>of</strong> de has nothing to do with the definiteness specification.Instead the major function <strong>of</strong> de is to associate the relative clause to the relative head. <strong>The</strong> function ispurely syntactic rather than semantic.150


4.3.2.3 Den Dikken & Singhapreecha (2004)Den Dikken and Singhapreecha (2004) analyzes de as a reflex <strong>of</strong> Predicate Inversion,explaining the presence <strong>of</strong> the marker and the predicative reading <strong>of</strong> the modifiers. For acomplex nominal like [XP de NP], the XP starts as the predicate <strong>of</strong> a small clause withthe NP being the subject. <strong>The</strong> XP predicate inverts around its NP subject via predicateinversion movement. Predicate inversion gives rise to the emergence <strong>of</strong> the complexfunctional head, and this complex head is spelt out as de.In the same spirit <strong>of</strong> Den Dikken and Singhapreecha, we propose that Chinese de,neither D nor C, is a complex functional head.de= Pr+F ([+Poss])In a possessive construction, F has the [+POSS] feature, thus inducing the possessionsemantics. With a (reduced) relative clause, F has the [-POSS] feature, thus unable toinduce the possession semantics. However, in both cases, de is the spell out <strong>of</strong> thecomplex head [Pr+F].(76) a.shubookzhangsanZhangsan151


.shubookyouquinterestingIn addition, [+POSS] F forms an accordance relation with the upper head, which is D.In Mandarin it is hard to see this accordance relation, since there is no definite article inMandarin. However, we can find evidence from Yi, which has the definite article:(77) mu gat ax yi ma suꃅꃅ ꀉ ꀉ ꂷ ꌠMuga child CL ART‘Muga’s that child’152


Pr (CL)ma i suCL ARTax yi t i mu gatchildMugaIn the derivation <strong>of</strong> (77), the classifier ma ꂷ adjoins to F, and then [CL+F] adjoinsto D. Yi differs from Mandarin in that the complex head [Pr+F] will not be spelt out as anew morpheme, because if Pr in Yi is overtly realized as CL, then the complex head[Pr+F] will be realized as CL, and if Pr is phonetically null, then the complex head [Pr+F]will not be spelt out. Rather it will invariably adjoin to D, and the complex head [Pr+F+D]will be spelt out as su. <strong>The</strong> complex head [Pr+F] in Yi always maintains an agreementrelation with the upper head D. If the whole nominal phrase is definite, then D will berealized as the definite article su, and the complex head [Pr+F] will adjoin to the definitearticle. If the whole nominal is indefinite, then the [-def] D will be null. In this case, wecan either assume that the complex head [Pr+F] adjoins to this null indefinite article, orassume that the complex head [Pr+F] stays in situ. In either way, there is an interaction153


etween [Pr+F] and D.4.4 SummaryIn this chapter, we have examined possessives in Mandarin and Yi. Yi has lexicalverbs to indicate possession. <strong>The</strong> most widely used verb is bbop (ꁨ). Some existentialverbs can also be used to indicate possession. jjo (ꐥ) is a versatile verb and it can expresslocation, possession, and existence. We elaborate that an existential possessive sentenceis derived from a nominal possessive with the possessor extraction. In a possessiveconstruction, F has the [+POSS] feature, thus inducing the possession semantics. With a(reduced) relative clause, F has the [-POSS] feature, thus unable to induce the possessionsemantics. In addition, [+POSS] F forms an accordance relation with the upper head,which is D. <strong>The</strong>refore, Yi [CL+F] complex can adjoin to the definite article su (ꌠ). Afterthat, the remnant FP raises to [Spec,DP]. <strong>The</strong> F [+ POSS] only contributes to thespecificity <strong>of</strong> the whole nominal, since it is natural that the possession relation normallypresupposes the existence <strong>of</strong> the possessee. We believe that D position in Yi can beoccupied either by the definite article su or an indefinite article, which has no phoneticrealization.154


CHAPTER 5 Relatives in Mandarin and Yi<strong>The</strong> relation between possessive constructions and relative clauses has beenstudied for decades. Evidence from various languages such as Mandarin, Cantonese,Lahu, Yi, and English suggests that there is an inalienable relationship betweenpossessives and relatives, and this idea seems to be shared among quite a few linguists.Dixon (1966) demonstrated a formal similarity in the marking <strong>of</strong> relative clause andgenitive constructions in Australian languages. Matis<strong>of</strong>f (1972) highlighted thisrelationship in Lahu. Larson (1991) argued that genitive DPs are generated in thesame position as relatives. Kayne (1994) elaborated this relation in English andRomance languages. Yap, Gu and Liu (2005) demonstrated the grammaticalization <strong>of</strong>nominalizers, which play a central role in relatives and possessives, in a large number<strong>of</strong> East Asian languages. <strong>The</strong> situation is further complicated by the recentobservation made by linguists working on classifier languages that in classifierlanguages, besides different nominalizers, classifiers can also serve as the possessivemarker and the relativizer (Matthews and Yip 1944, 2001). This naturally leads us toreconsider the function <strong>of</strong> classifiers.Normally, the classifier is assumed to be a grammatical means for the linguisticcategorization <strong>of</strong> nouns (to be more accurate, nominals). Corresponding to differentnouns, classifiers play measuring or individuating function. How can classifiers gainthe function in relatives and possessives is a mystery. In the previous chapter onpossessives, we have demonstrated that classifiers can serve as Pr, helping form a155


predication relation within a SC. In this chapter we are going to extend the CL-as-Pranalysis to the relatives in Mandarin and Yi.5.1 Two approaches to relative clauses<strong>The</strong>re are mainly two different approaches to relative clauses in the currentliterature. <strong>The</strong> first one is called matching approach, and the second one is calledraising/promotion approach.5.1.1 Matching approachChomsky (1977) proposes that, like wh-interrogatives, relative clauses arederived via operator movement (as are clefts, topicalizations, easy-to-pleaseconstructions, etc.)<strong>The</strong> Matching Analysis:[ NP/DP [Head NP/DP i …] [ Relative CP wh i [ IP … t i …]]]Under the matching analysis, the Head is base-generated. A wh-operator is moved to aposition close to the Head and bears a predication or agreement relation to the Head,marked by the co-index. <strong>The</strong> relative CP adjoins to the relative Head. 11 Such a right adjoined structure is not allowed in Kayne’s LCA. Refer to Kayne (1994: chapter 3) forthe illustration why such a right adjunction is not allowed according to Kayne’s LCA.156


5.1.2 Raising/promotion approachThis approach is different from the matching approach in that under thisapproach the relative clause does not adjoin to the relative head. Rather, either somefunctional head is raised from a lower position to a higher position, or the relativehead is raised from inside the relative clause. This approach is represented byLarson’s (1991) D-to-D raising analysis and Kayne’s (1994) D-CP analysis.5.1.2.1 Larson (1991)Larson (1991) approaches the DP phrase structure from the semantics <strong>of</strong>determiners based on generalized quantifier theory. He argues that research inquantification theory has yielded a genuinely independent characterization <strong>of</strong>argument structure for determiners (D). Determiners express quantification, andnotions like restriction and scope represent two main semantic roles. Semantically, therestriction sets the domain <strong>of</strong> quantification, whereas the scope determines what istrue <strong>of</strong> those individuals in the domain <strong>of</strong> quantification. Syntactically, restriction andscope are also plainly relevant in mapping the parts <strong>of</strong> DP. <strong>The</strong> former role is mappedto the NP complement <strong>of</strong> D. <strong>The</strong> latter role is associated with a main clausepredication. Based on these assumptions, he proposes the projection <strong>of</strong> DP:(1)157


Larson suggests that the semantic value <strong>of</strong> the Pro argument is determinedconfigurationally at the level <strong>of</strong> Logical Form. Specifically Pro gets its value from thederived predicate that is the structural sister <strong>of</strong> DP at LF. Based on this general picture<strong>of</strong> the projection <strong>of</strong> DP, Larson proposes the structure <strong>of</strong> relative clauses and genitives.He gives the following data to illustrate that RCs are more appropriately analyzable ascomplement <strong>of</strong> D.(2) I earned it that way/*the way/the old-fashioned way/the way that one should.What (2) shows is the discontinuous dependency holding between the determiner andthe restrictive modifier, whether relative clause, attributive adjective or PP. <strong>The</strong>determiner cannot survive without them. Larson proposes the following analysis toaccount for the dependency in terms <strong>of</strong> selection between D and its sister modifier CP.(3)Larson extends such an analysis to possessives like John’s briefcase, where thepossessor John and the possessee briefcase are both arguments <strong>of</strong> a definite158


determiner THE, which raises up:(4) a. [ DP Pro e [ DP John’s [D’ THE briefcase]]]b. [ DP Pro THE [ DP John’s [D’ t briefcase]]]|_______________ |Larson argues that (4) is similar to his analysis <strong>of</strong> DOC (Larson 1988):(5) a. [ VP Mary e [ VP John [V’ gave a briefcase]]]b. [ VP Mary gave [ VP John [ V’ t a briefcase]]]|_______________|Larson analyzes genitive nominals as the DP-equivalents <strong>of</strong> double objectconstructions in the verbal domain. What is interesting about Larson’s analysis is thathe manages to unify the generation <strong>of</strong> possessives and relatives by the samemovement strategy: the movement <strong>of</strong> D. This is quite illuminating as it helps explainwhy in many languages genitive morphology is the same with the relativemorphology: they are derived by the same syntactical process. Larson’s analysis hastwo appealing points. <strong>The</strong> first is that he successfully explains the observation thatdeterminers and relative clauses form a constituent. <strong>The</strong> second is that he successfullyexplains the identity <strong>of</strong> genitive morphology and relative morphology in manylanguages.In short, Larson’s analysis can be called D-raising analysis. For relative clauses,they form a constituent with D. This is also in conformity with Kayne’s “D-CP”hypothesis. What differentiates Larson’s analysis from Kayne’s is that Larson assumestwo layers <strong>of</strong> DP, with the lower D promoted to the upper D, in the same manner thatV moves to v in verbal domain.159


5.1.2.2 Kayne (1994)Kayne (1994) also approaches relatives together with possessives. FollowingSzabolcsi’s (1981) analysis <strong>of</strong> Hungarian possessives, Kayne argues that it isadvantageous to take English to have a phonetically unrealized counterpart to theHungarian D.(6) John’s two pictures(7) a. D [ FP John [’s [two pictures]]]b. [ DP --D [ FP NP1 F [NP2]]]<strong>The</strong> possessor (NP1) John resides at [Spec,FP]. If we move NP2, then we get:(8) [ DP [two pictures] i [ D <strong>of</strong>] [ FP John [’s [e] i ]]]D in (7b) is occupied by the preposition <strong>of</strong> in (8). Kayne argues that D cannot have aDP complement, so (9) is not acceptable.(9) *I found the two pictures <strong>of</strong> John’s.<strong>The</strong> most important observation used by Kayne to elaborate his analysis <strong>of</strong> relatives isthat although (9) is out, (10) is perfect.(10) I found the two pictures <strong>of</strong> John’s that you lent me.Kayne argues that the two pictures <strong>of</strong> John’s is not a constituent. Instead two pictures<strong>of</strong> John’s that you lent me is a constituent. It is not a DP, but a CP. two pictures <strong>of</strong>John’s is at [Spec,CP] as the result <strong>of</strong> a movement. In this way, Kayne comes to theconclusion that the raising/promotion analysis <strong>of</strong> relative clauses should hold.160


According to Kayne (1994), the promotion analysis involves the followingcomplementation structure and the Head movement process:(11) <strong>The</strong> Promotion Analysis: [D CP][ DP D [ CP NP i [C [ IP …t i …]]]]<strong>The</strong> above derivation is for English relatives. Under this analysis, there is a positioncompetition between the relative head and the relative pronoun, since both <strong>of</strong> themoccupy the position [Spec,CP]. For example:(12) the book which I likeIn (12) both book and which should occupy the position [Spec, CP]. Kayne solves thisproblem by proposing that which is a wh-determiner D, whose complement is book.<strong>The</strong> derivation <strong>of</strong> (12) is as follows:(13) the [ CP which book [C…(14) the [ CP book which [C….<strong>The</strong> relative head ‘book’ use the specifier position <strong>of</strong> the wh-determiner as a landingsite. (13) and (14) show the derivational process <strong>of</strong> N-initial relative clauses.In many languages, the relative head follows the relative clause (N-final relatives).Kayne focuses on Amharic to illustrate the derivation process <strong>of</strong> N-final relatives. Inthis language, the relative clause precedes the definite article, which itself precedesthe relative head noun (RC>D>NP). Kayne argues that the word order suggests thatthe relative clause has moved into [Spec,DP]. To explain why the relative headfollows D, Kayne proposes that what has moved to [Spec,DP] is not the whole CP.161


Rather it is the IP. Recall the structure given for English relatives:(15) the [[ NP picture] i [that [ IP Bill saw [e] i ]]]Here the NP picture has moved to [Spec,CP], leaving a trace in object position.Moving IP to [Spec,DP], we can get (16):(16) IP j the [[picture] i that t j ]Hence the relative clause is actually the moved IP. For Amharic, the complementizerthat is invisible.Kayne also mentions that internally headed relatives are a kind <strong>of</strong> N-finalrelatives. <strong>The</strong> difference between the internally headed relatives and the normalN-final relatives lies in which instance <strong>of</strong> the N, either the tail or the head <strong>of</strong> therelative N, is phonetically realized. For the internally headed relatives, the tail isrealized. For the normal N-final relatives, the head is realized. Recall the analysisgiven by Kayne for N-final relatives:(17) IP j D [[picture] i C t j ]<strong>The</strong> IP contains a trace <strong>of</strong> picture, which has already moved to [Spec,CP]:(18) [[ IP ……[picture] i ….] j D [[picture] i C IP j ]For internally headed relatives, the second instance <strong>of</strong> picture is deleted in PF. Fornormal N-final relatives, the first instance is deleted.162


5.1.3 SummaryChomsky’s (1977) matching analysis <strong>of</strong> RC has not taken the special dependencybetween D and RC into consideration, and the right adjunction <strong>of</strong> RC to the relativehead will give rise to indeterminacy <strong>of</strong> the linear word order according to LCA.<strong>The</strong>refore, this approach will be rejected in this thesis. However, we keep the spiritthat there is a predication relation holding between the relative head the relativeclause.Larson’s analysis <strong>of</strong> viewing relatives and possessives as two-layer DPs capturesthe dependency between D and relatives and possessives. For relative clauses, withinthe lower DP, we have the word order [Relative head—D—RC]. If we shift data fromarticle languages like English to classifier languages like Cantonese, we can readilyget [Relative head—CL—RC]. 2 For languages like Yi which has both CL and definitearticle, we may assume that the lower D is realized as CL, whereas the higher D isoccupied by the definite article.Kayne’s analysis is powerful in terms <strong>of</strong> RC typology. In the following sectionwe will test whether Kayne’s raising analysis can be applied to N-final relatives <strong>of</strong>classifier languages like Mandarin which has no article and N-initial relatives <strong>of</strong>classifier languages like Yi which has article.2 <strong>The</strong> correct word order <strong>of</strong> Cantonese RC with classifiers is [RC-CL-Relative head], which can beargued to be derived from the underlying word order [Relative head-CL-RC] via predicate inversionMove.163


5.2 Relative clauses in MandarinIn this section, we will first review two influential analyses <strong>of</strong> Mandarin relativeclauses in literature. After that we will give our CL-as-Pr analysis <strong>of</strong> Mandarinrelatives.5.2.1 Two approaches to Mandarin relatives5.2.1.1 Simpson (2002)Simpson <strong>of</strong>fers a D-CP analysis for Chinese relatives by claiming de is D.Simpson’s analysis <strong>of</strong> Chinese relatives is identical to Kayne’s treatment <strong>of</strong> Amharicrelatives, since Simpson views de as D. 3(19)deDEwo mai shuI buy book3 However, it is quite questionable to treat de as a determiner, since de does not possess the typicalcategorial features <strong>of</strong> determiners both semantically and syntactically.164


5.2.1.2 Aoun and Li (2003)Aoun and Li (2003) argue that different types <strong>of</strong> relative constructions requiredifferent empirical generalizations. Relatives differ with regard to whether the relativehead can be reconstructed. <strong>The</strong>y argue that there are three strategies to get relatives:(20) a. a promotion strategy that moves the Head <strong>of</strong> the relative constructionb. an operator movement strategyc. a direct base-generation strategy where movement is not available.English relatives can be classified into two types: those derived by Head raising andthose derived by operator movement. <strong>The</strong> latter is realized as wh-relatives and theformer as non-wh-relatives.(21)<strong>The</strong> NP at [Spec,ForceP] is base-generated. What occupies [Spec,TopP] is awh-operator. Unlike Kayne’s analysis, this wh-operator does not have a lexical NP asits complement.165


(22)<strong>The</strong> choice <strong>of</strong> derivation for relative constructions depends on what is generated asthe phrase to be relativized: either a DP with an empty D or a wh-DP. (21) and (22)illustrate the derivational process <strong>of</strong> Head-initial relatives. For Head-final relatives,Aoun and Li claim that evidence shows that relative constructions in Chinese involveadjunction structures and NP movement. Thus Aoun and Li conclude that Head-initialand Head-final relatives can be reduced to a difference in their phrase structure,complementation versus adjunction.5.2.2 Mandarin restrictive and non-restrictive relativesIt has been long noticed that in Chinese there exist two kinds <strong>of</strong> relatives. Li &Thompson (1981: 124) propose that the order <strong>of</strong> elements in a noun phrase has thefollowing two schemas (23) and (24), exemplified in (25) and (26):166


(23) associative phrase+classifier/measure phrase+relative clause+adjective+noun(24) associative phrase+ relative clause+ classifier/measure phrase +adjective+noun(25) wo mai de na ben shu我 买 的 那 本 书I buy DE that CL book‘the book that I bought’(26) na ben wo mai de shu那 本 我 买 的 书that CL I buy DE book‘that book, which I bought’According to Chao (1968), whether a relative in Chinese is restrictive ordescriptive (non-restrictive) depends on the position <strong>of</strong> the demonstrative modifier inthe sentence. If the relative follows a demonstrative, it is descriptive, as is shown in(26); the function <strong>of</strong> the relative clause is to describe the head noun rather thandistinguish it from others, and the relative is restrictive, if it precedes a demonstrative.Huang (1982: 69) gives a detailed analysis <strong>of</strong> the two types <strong>of</strong> Mandarin relativesand has changed the terminology Descriptive into Non-restrictive, which explicitlymeans that a non-restrictive clause has no ability to specify the reference. He employsScope <strong>The</strong>ory to explain the two different types <strong>of</strong> nominal structures. With thedemonstrative outside <strong>of</strong> the scope <strong>of</strong> the relative, the relative modifies only the head167


noun but not the demonstrative. <strong>The</strong> demonstrative, whose referential function is notunder the effect <strong>of</strong> any modifier, is used deictically. <strong>The</strong> relative clause following ittherefore does not need to participate in the determination <strong>of</strong> the NP’s reference. Sincethe reference is already sufficiently determined by the deictic demonstrative, therelative clause has only a descriptive function. If the demonstrative is within the scope<strong>of</strong> a relative clause, the referential value is subject to the c-commanding relativeclause. <strong>The</strong> gap within the relative clause functions as a variable bound by the relativehead. Huang’s argument specifies different sources <strong>of</strong> definiteness <strong>of</strong> each sentence,but he does not specify in detail how the relative relates to the value <strong>of</strong> thequantificational N binding the variable; i.e. how the referent <strong>of</strong> the head noun issubject to the c-commanding relative clause.Adopting the antisymmetry model, we can claim that the underlying form <strong>of</strong> therelative clauses (25) and (26) could be (27) and (28).(27) D [ CP C[ IP wo mai [ DP na ben shu]]] 4(28) [ IP wo mai shu]In (27) the DP naben shu (‘that book’) is first raised to [Spec,CP], and then theIP-RC is raised to [Spec,DP]. <strong>The</strong> definiteness <strong>of</strong> (27) comes from the demonstrativeimmediately preceding the noun. Following the same model, (28) will generate arelative clause womai de shu (‘book that I bought’), and then it merges to the‘Demonstrative+Classifer’ construction, and gains the definiteness from the4 Although in the antisymmetry model the element that is raised to Spec,CP is normally an NP, Kayne(1994: 158, note 26) argues that an indefinite DP can also be promoted to that position, whereas adefinite DP would be impossible. In Mandarin even a definite DP like naben shu ( 那 本 书 ) is alsopermissible in that position.168


demonstrative. In both cases the definiteness <strong>of</strong> the expression comes from the deicticfunction <strong>of</strong> the demonstrative. de is proven to have no function in determining thedefiniteness <strong>of</strong> the nominal phrase. It only functions to link the relative clause and therelative head.5.2.2 CL-as-Pr analysis <strong>of</strong> Mandarin relativesIn this section, we will propose our analysis <strong>of</strong> Mandarin relatives exemplified in(25) and (26), repeated here as (29) and (30):(29) wo mai de na ben shu我 买 的 那 本 书I buy DE that CL book‘the book that I bought’(30) na ben wo mai de shu那 本 我 买 的 书that CL I buy DE book‘that book, which I bought’Based on Huang’s observation <strong>of</strong> the different scope <strong>of</strong> demonstratives, wepropose that (29) and (30) have undergone different derivational process. For (29), therelative head is the CLP na ben shu ( 那 本 书 ). <strong>The</strong> derivation <strong>of</strong> this phrase is169


illustrated by the following tree:(31)shu ben nabook CL thatAssuming that Kayne’s D-CP hypothesis can also be applied to Mandarin, 5 andthat our CL-as-Pr analysis is also on the right track, we can get the followinggeneration.(32)[na ben shu] i[that CL book] it iwo mai t iI buy t i5 A minor modification is needed here. In classifier languages the CP is not the complement <strong>of</strong> D.Rather the CP is the complement <strong>of</strong> CL.170


What has been raised to [Spec, CP] is the FP na ben shu ( 那 本 书 ), which furthermoves up to [Spec, CLP]. At this stage, we can get the predication relation betweenthe phrase na ben shu ( 那 本 书 ) and the IP with a gap wo mai t ( 我 买 t). <strong>The</strong>predication relation is established by the empty CL. This empty CL will incorporateinto F higher up, and enable the predicate inversion Move. <strong>The</strong> complex functionalhead [CL+F] surfaces as de. <strong>The</strong> reason why CL in this case has to be empty is due tothe fact that the phrase at [Spec, CLP] na ben shu ( 那 本 书 ) already contains a CL ben( 本 ). 6 <strong>The</strong> derivation <strong>of</strong> the phrase (30) is illustrated by the following trees:(33)[shu] i[book] iwo mai t iI buy t i6 In Cantonese, but not in Mandarin, if the phrase at [Spec, CLP] is a bare noun, then CL willphonetically realized, and it will incorporate into the empty F. <strong>The</strong> complex [CL+F], surfacing as CL,will serve as the relative clause marker. Matthews and Yip (1994, 2001) divide Cantonese relatives intotwo types: relatives with ge, and relatives with classifier. <strong>The</strong>y argue that the two types <strong>of</strong> relativesbelong to different strata. Relatives with classifier is extremely common, being preferred in colloquialCantonese to the more formal relatives with ge. We would propose that when CL is overt, then [CL+F]will surface as CL; whereas when CL is empty, then [CL+F] will surface as ge in Cantonese.171


We first get wo mai de shu ( 我 买 的 书 ). This phrase then serves as the subject <strong>of</strong>the whole predication phrase:(34)wo mai de shu ben naI buy DE book CL that<strong>The</strong> above analysis seems to be on the right track, and further supports Kayne’spromotion analysis. However, there is a fatal mistake in the derivation, which forcesme to abandon Kayne’s promotion analysis, and pick up the matching analysis <strong>of</strong>relative clauses. In (32) the relative head is raised out from the argument position to[Spec, CP]. This is a typical A’ movement from an argument position to anon-argument position. However, [Spec, CP] is just an intermediate landing site. Inorder to capture the predication relation, the relative head has to land in [Spec, CLP]position, which is a subject position. <strong>The</strong> chain has three links A-A’-A. This is fatal,since the relative head assumes two theta roles, which violates theta-criterion. If we172


assume that the relative head is base-generated at [Spec, CLP], and the gap in therelative clause is a variable, bound by an operator at [Spec, CP], then the problem issolved. <strong>The</strong> advantage <strong>of</strong> taking the matching analysis is that it makes possible thepredication relation holding between the relative head and RC; one the other hand, iteludes derivational problem found in (32). <strong>The</strong> difference between English andclassifier languages is that the predication relation is marked by co-indexation inEnglish, but the predication relation is established by a functional head CL/Pr inclassifier languages.5.2.3 Analysis <strong>of</strong> Reduced RelativesAccording to Cinque (2005 LSA handout), there is a universal structure <strong>of</strong> DPsbased on his research on the two sources <strong>of</strong> adjectives in Romance languages andGermanic languages. He argues that adnominal adjectives have two separate sources.One is a direct adnominal modification source, which he takes to involve merge <strong>of</strong> thedifferent classes <strong>of</strong> APs in the specifier <strong>of</strong> various dedicated functional heads <strong>of</strong> theextended projection <strong>of</strong> the NP. <strong>The</strong> other is a relative clause source. <strong>The</strong> lattercorresponds to Sproat and Shih’s (1988, 1991) “indirect modification”. Cinque alsoargues that the merge position <strong>of</strong> relative clauses is always prenominal, specifically inthe specifier <strong>of</strong> a projection above the projections hosting the direct adnominalmodifier APs, as is very roughly sketched in the following:173


(35)<strong>The</strong> postnominal position <strong>of</strong> (reduced) relative clauses will be taken to arise as aconsequence <strong>of</strong> the RC-IP raising to the specifier <strong>of</strong> higher C followed by the merge<strong>of</strong> an (overt or covert) complementizer that attracts the remnant along lines advocatedby Kayne (1994).Chinese de can occur with reduced relative clauses. Besides possessives andrelatives, de can also occur with pre-nominal adjective phrases and pre-nominalprepositional phrases. Traditionally, it is believed that de in these constructions hasdifferent functions from that in possessives and relatives. However, after comparingpre-nominal adjective phrases and pre-nominal prepositional phrases with relatives,we can reach a unified explanation for these three constructions. In the following wewill examine their syntactic structures respectively.174


5.2.3.1 De with Pre-nominal Adjective PhrasesSince adjectives are intrinsically predicate, there should be no difference betweenverbs and adjectives with their syntactic distribution. 7 However, English adjectiveshave to co-occur with a copular verb to realize the predicative function. Chinese doesnot need copular verbs. Adjectives can be directly used as predicates. 8(36) fangzi hen da.房 子 很 大 。housevery big‘<strong>The</strong> house is big.’(37) hen da de fangzi很 大 的 房 子very big DE house‘a house that is very big’7 Chinese adjectives can directly go with the head noun without de in between. We can either adoptCinque’s analysis that adjectives are at [Spec, FP] with F being various dedicated functional heads, oradopt the predicate inversion analysis. In this thesis, we adopt the latter.8 Gu (2006) points out that hen ( 很 ) is important in the sense that it locates the following adjectivealong a scale. Thus the predicate <strong>of</strong> the sentence is marked deictic. This kind <strong>of</strong> deicticity marking canbe viewed as a kind <strong>of</strong> generalized tense/aspect (INFL) anchoring.175


We can argue that (37) contains a reduced relative clause, whose underlying structureis (36). Based on the matching analysis, the relative head is base-generated at [Spec,CLP]. <strong>The</strong>re is an operator at [Spec, CP] binding the variable in the RC. <strong>The</strong>movement involved is predicate inversion, and the resultant complex head [CL+F]surfaces as de. <strong>The</strong> derivation <strong>of</strong> (37) is shown in (38).(38)[fangzi] iOP i[house] it i hen dat i very big176


5.2.3.2 De with Pre-nominal Preposition PhrasesIn Mandarin, some PPs can also serve as predicates. For example,(39) shu zai zhuo shang.书 在 桌 上 。book at desk top‘<strong>The</strong> book is on the desk.’We also adopt the matching analysis to explain the generation <strong>of</strong> the pre-nominalpreposition phrases. <strong>The</strong> derivation <strong>of</strong> (40) is shown in (41).(40) zai zhuo shang. de shu在 桌 上 的 书at desk top DE book‘the book on the table’(41)OP i[shu] i[book] it i zai zhuo shangt i at desk top177


Notice that this analysis does not hold for every PP in Mandarin, because thereare some pseudo- predicate PPs in Mandarin. Compare examples (42) and (43):(42) Meiguo dui Yilaka de taidu bu manyi美 国 对 伊 拉 克 的 态 度 不 满 意America to Iraq DE attitude not satisfy‘America was not satisfied with Iraq’s attitude’(43) Meiguo dui Yilaka de taidu bu hao美 国 对 伊 拉 克 的 态 度 不 好America to Iraq DE attitude not good‘America’s attitude towards Iraq is not good’(42) and (43) illustrate the structural ambiguity <strong>of</strong> the string Meiguo dui Yilakade taidu ( 美 国 对 伊 拉 克 的 态 度 ). In (42) the scope <strong>of</strong> the preposition is the DP Yilakade taidu ( 伊 拉 克 的 态 度 ), whereas in (43) the scope <strong>of</strong> the preposition is the DPYilaka (‘Iraq’). <strong>The</strong> two readings <strong>of</strong> this string can be differentiated by the mainpredicate <strong>of</strong> the sentence. For (42) the predicate is a verb, and the subject is Meiguo(‘America’), and the meaning <strong>of</strong> the sentence is identical to that <strong>of</strong> (44):(44) Meiguo bu manyi Yilaka de taidu.美 国 不 满 意 伊 拉 克 的 态 度America not satisfy Iraq DE attitude‘America was not satisfied with Iraq’s attitude.’178


<strong>The</strong> insertion <strong>of</strong> the preposition <strong>of</strong> dui (‘to’) enables the object DP to be raised tothe left <strong>of</strong> the verb. For (43) the predicate is an adjective, and the subject is taidu(‘attitude’). In both cases, the pre-nominal PP cannot serve as a predicate.5.3 Relatives in YiYi has the definite article su, which can go with relative clauses. Yi classifierscan also mark relative clauses. In this section, we will examine three different types <strong>of</strong>relatives in Yi.5.3.1 Three types <strong>of</strong> relatives in YiYi relatives can be divided into three different types shown in (45) andexemplified in (46) to (48):(45) a. Relatives with classifierb. Relatives with suc. Relatives with classifier+su(46) vot mu gat ho da ma at gop gep sit ox.ꃮ ꃅꃅ ꉻ ꄉ ꂷ ꀈꀈ ꇱ ꌉ ꀐ.pig Mu Ka feed ASP CL At Gop AGENT kill ASP‘At Gop killed a pig that Mu Ka fed.’179


(47) vot mu gat ho da su at gop gep sit ox.ꃮ ꃅꃅ ꉻ ꄉ ꌠ ꀈꀈ ꇱ ꌉ ꀐ.pig Mu Ka feed ASP ART At Gop AGENT kill ASP‘At Gop killed pigs that Mu Ka fed.’(48) vot mu gat ho da ma su at gop gep sit ox.ꃮ ꃅꃅ ꉻ ꄉ ꂷ ꌠ ꀈꀈ ꇱ ꌉ ꀐ.pig Mu Ka feed ASP CL ART At Gop AGENT kill ASP‘At Gop killed that pig that At Gop fed.’<strong>The</strong> three different types <strong>of</strong> relatives differ significantly in terms <strong>of</strong> referentialproperty <strong>of</strong> the relative head. For the first type, as in (46), the relative head isindefinite singular; for the second type, as in (47), the relative head is definite singularor plural; for the third type, as in (48), the relative head is definite singular. It can besummarized as follows:Table 1: Referential property <strong>of</strong> the relative head in three types <strong>of</strong> Yi relativesTypes Definiteness Number feature <strong>of</strong> the relative headi) Relatives with classifier Indefinite Singularii) Relatives with su Definite Singular/Pluraliii) Relatives with classifier + su Definite Singular180


5.3.2 Relative word order <strong>of</strong> Yi relativesYi is a Relative-Head-initial (N-initial) language like English, but unlike English,the definite article su occurs at the postnominal position. <strong>The</strong> following chart gives asummary <strong>of</strong> the relative word order <strong>of</strong> relative clauses in four typologically differentlanguages.Table 2: <strong>The</strong> relative word order between RC, D, and N in four languagesLanguage Relative head N-position Overall word orderEnglish N-initial D>N>RCYi N-initial N>RC>DAmharic 9 N-final RC>D>NMandarin N-final RC>de>NIn English and Yi, the relative head is before the relative clause. <strong>The</strong>ir differencelies in the position <strong>of</strong> D. In English, D precedes the relative head; whereas in Yi it is atthe final position.In Amharic and Mandarin, the relative head is after the relative clause. <strong>The</strong>irdifference lies in the fact that there is a D in Amharic which occurs between therelative clause and the relative head; whereas in Mandarin, it is a complex functionalhead de which occurs between the relative clause and the relative head.9 We follow Kayne’s (1994: chapter 8) description <strong>of</strong> Amharic.181


5.3.3 Yi reduced relativesIn Yi, adjectives cannot directly modify nouns unless they have formed aconstituent with classifiers or su. <strong>The</strong>y simply cannot stand alone. For example:(49) *co o bbu la ox.ꊿ ꀑꀑ ꇁ ꀐperson clever come ASP(50) co o bbu ma la ox.ꊿ ꀑꀑ ꂷ ꇁ ꀐperson clever CLcome ASP‘A clever person came.’(51) co o bbu su la ox.ꊿ ꀑꀑ ꌠ ꇁ ꀐperson clever ART come ASP‘(All) the clever people came.’(49) is ungrammatical. This serves as evidence that Yi adjectives cannot be usedattributively. <strong>The</strong>y can only be used predicatively. This is very crucial for thefollowing discussion, because this shows that Yi adjectives cannot functionattributively. <strong>The</strong>y can only be used as predicative adjectives.182


Following Cinque’s (2005) observation <strong>of</strong> two sources <strong>of</strong> adjectives, Yi naturallyfalls into the category <strong>of</strong> languages lacking attributive (direct modification) adjectives.(50) differs from (51) in that (50) specifies that the grammatical number <strong>of</strong> the headnoun is singular, whereas (51) does not indicate how many people came. Sinceadjectives are argued to have only predicative function, we are forced to say that (50)and (51) involve reduced relative clauses. Adopting our CL-CP analysis, we wouldpropose the following analysis for Yi reduced relatives. For (50) we have thefollowing tree diagram to illustrate the needed derivational process:(52)OP i183co i maperson i CLt i obbut i clever


<strong>The</strong> relative head is base-generated at [Spec, CLP]. <strong>The</strong> predication relationbetween the relative head and the relative clause is established by the head CL (Pr).CL will incorporate into F, and the remnant CLP will undergo A’-movement to[Spec,DP].For (51) we have the following tree diagram to illustrate the needed derivationalprocess:(53)OP i184suARTco iperson it i obbut i clever


In this case, the definite article su occupies D. <strong>The</strong> predication relation betweenthe relative head and the relative clause is established by the head CL (Pr), which hasno phonetic realization. Since there is no classifier appearing in the CLP, the CLPdenotes a whole set (the set <strong>of</strong> all smart people for (51)). This whole set is unique andcannot be partitioned. <strong>The</strong>refore, it can move to [Spec,DP] to check the [+def] feature<strong>of</strong> the definite article su.If we do not denote the whole set. Rather we would want to specify a particularindividual in the set. We can have the following sentence in Yi.(54) co o bbu ma su la ox.ꊿ ꀑꀑ ꂷ ꌠ ꇁ ꀐperson clever CL ART come ASP‘<strong>The</strong> clever person came.’In this sentence, the subject is a definite nominal containing both classifier andthe definite article, and it refers to a unique definite clever person. <strong>The</strong> derivationalprocess is as follows:185


(55)OP isuARTco i maperson i CLt i obbut i cleverIn this case, CL will first adjoin to F, and the complex head [CL+F] furtheradjoins to D. <strong>The</strong> CLP is raised to [Spec, DP] to check the [+def] feature <strong>of</strong> D.186


5.3.4 An analysis <strong>of</strong> Yi relativesTypical relative clauses <strong>of</strong> Yi are exemplified in (56) to (59):(56) nga li cy vy su ka.ꉢ ꆹ ꋌ ꃼ ꌠ ꈁI TOP he buy ART want‘I want what he bought.(57) co bbo su lax la ox?ꊿ ꁧ ꌠ ꇀ ꇁ ꀐperson go ART come come ASP‘Did the person who went there return?’(58) yo ry zze su tep la ox.ꑿ ꏜ ꋠ ꌠ ꄯ ꇁ ꀐ.sheep grass eat ART alive come ASP‘<strong>The</strong> sheep who ate grass came to life again.’(59) vot mu gat ho da su at gop gep sit ox.ꃮ ꃅꃅ ꉻ ꄉ ꌠ ꀈꀈ ꇱ ꌉ ꀐ.pig Mu Ka feed ASP ART At Gop AGENT kill ASP‘At Gop killed pigs that Mu Ka fed.’187


In Yi, su is the definite article, occupying D. All the material following D has tomove to [Spec,DP]. <strong>The</strong> derivation is as follows:(60)OP iSo far we have only seen relatives with the definite article su. In fact, the definitearticle su can be optional. When this happens, classifier will take the function <strong>of</strong>forming relatives. However, the relative head becomes immediately indefinite.188


(61) vot mu gat ho da ma at gop gep sit ox.ꃮ ꃅꃅ ꉻ ꄉ ꂷ ꀈꀈ ꇱ ꌉ ꀐ.pig Mu Ka feed ASP CL At Gop AGENT kill ASP‘At Gop killed a pig that Mu Ka fed.’Since classifies by default carries the singular grammatical feature, the relativehead thus gets the singular interpretation. If a relative head happens to be singular anddefinite, then we will have a more complicated relative structure. (62) contains both aclassifier and the definite article to mark the singular definite relative head.(62) vot mu gat ho da ma su at gop gep sit ox.ꃮ ꃅꃅ ꉻ ꄉ ꂷ ꌠ ꀈꀈ ꇱ ꌉ ꀐ.pig Mu Ka feed ASP CL ART At Gop AGENT kill ASP‘At Gop killed that pig that At Gop fed.’To sum up, Yi relatives have three types according to different semantic content<strong>of</strong> the relative head shown in (59), (61) and (62), repeated here as (63)-(65):(63) vot mu gat ho da ma at gop gep sit ox.ꃮ ꃅꃅ ꉻ ꄉ ꂷ ꀈꀈ ꇱ ꌉ ꀐ.pig Mu Ka feed ASP CL At Gop AGENT kill ASP‘At Gop killed a pig that Mu Ka fed.’189


(64) vot mu gat ho da su at gop gep sit ox.ꃮ ꃅꃅ ꉻ ꄉ ꌠ ꀈꀈ ꇱ ꌉ ꀐ.pig Mu Ka feed ASP ART At Gop AGENT kill ASP‘At Gop killed pigs that Mu Ka fed.’(65) vot mu gat ho da ma su at gop gep sit ox.ꃮ ꃅꃅ ꉻ ꄉ ꂷ ꌠ ꀈꀈ ꇱ ꌉ ꀐ.pig Mu Ka feed ASP CL ART At Gop AGENT kill ASP‘At Gop killed that pig that At Gop fed.’<strong>The</strong> relative head in (63) is singular indefinite; the relative head in (64) isdefinite with the number feature unspecified; the relative head in (65) is singular anddefinite. <strong>The</strong> relative heads in (63) and (65), that is relatives with classifiers, can beturned into plural by adding cardinal numbers before the classifiers.We adopt Larson’s idea that the structure <strong>of</strong> relatives and genitives involves atwo DP-layer structure and we propose that for classifier languages, relatives andpossessives both have a lower CLP-layer and a higher DP layer. <strong>The</strong> detailedderivation process for Yi full relative clauses is identical to that <strong>of</strong> Yi reducedrelatives.190


5.4 SummaryIn this chapter we have examined the syntax <strong>of</strong> relatives in Mandarin and Yi.Since Mandarin has no definite article to occupy the D position, we propose that thestructure <strong>of</strong> Mandarin relatives involves two layers <strong>of</strong> predication relations. One <strong>of</strong>them is to form the predication relation between the relative head and the relativeclause, and the other is to determine the referential property <strong>of</strong> the head noun.Yi has the definite article to occupy the D position, and we propose that Yi has alower CLP layer and a higher DP layer. <strong>The</strong> CLP layer functions to form thepredication relation between the relative head and the relative clause; whereas the DPlayer functions to determine the referential property <strong>of</strong> the relative head.191


CHAPTER 6 <strong>Complex</strong> nominals in Jingpo: a case in focusIn previous Chapters we have already applied the CL-as-Pr analysis topossessives and relatives both in Mandarin and Yi. <strong>The</strong> difference between Mandarinand Yi lies largely on the fact that Yi has the definite article su; whereas Mandarinlacks the article system. 1 Definite complex nominals in Mandarin contain DemPs andclassifiers. Despite the difference between Mandarin and Yi, we can see that the basicstructure for possessives and relatives is almost the same. In both languages, the basicstructure for possessives and relatives is as follows:(1)<strong>The</strong> whole CLP is a PrP, with XP serving as the subject <strong>of</strong> the predicationrelation, and YP serving as the predicate. <strong>The</strong> possessive construction and relativeclause construction can be briefly illustrated by the following trees:1 Both Cheng and Sybesma’s (1999) analysis and Li’s (1999) analysis fail to deal with the nominalstructure with overt Determiner in classifier languages like Yi. Cheng and Sybesma’s analysis does notassume the projection <strong>of</strong> D, since the definiteness encoding can be dealt with by ClP. Li’s analysisargues for the projection <strong>of</strong> D in Mandarin. However, under her analysis D can only be occupied byproper names, personal pronouns and common nouns which have undergone N-to-D movement. Withthe NumP between DP and CLP, her analysis cannot deal with classifier languages with overtDeterminer like the definite article su in Yi.192


(2) a.b.In this chapter we will apply this CL-as-Pr analysis to Jingpo possessives andrelatives. Like Yi, Jingpo is also a representative language in the Tibeto-Burmanlanguage family. It is a classifier language with SOV word order.6.1 Demonstratives in Jingpo<strong>The</strong> demonstrative system <strong>of</strong> Jingpo is unique in that there are altogether fivedemonstratives in the language, with two <strong>of</strong> them being the usual proximal one ndai(this) and the distal one dai (that), while the remaining three are relevant to anotherdeictic dimension, namely elevation. In Jingpo, there are two plural morphemes -niand -hte, and they can be attached to all five demonstratives to express plurality.According to Cheung (2003), demonstratives inflected with the two plural morphemescan denote different types <strong>of</strong> referents: more precisely, the appearance <strong>of</strong> Dem + -hte193


can only refer to non-human entities, whereas Dem + -ni is used to denote humanreferents.Table 1: Demonstratives in Jingpo 2Dem -ni (human)Dem -hte (non-human)Proximal Ndai-ni Ndai-hteDistal Dai-ni Dai-hteUp Htora-ni Htora-hteLevel Lera-ni Lera-hteDown Wora-ni Wora-hteWith respect to the distribution <strong>of</strong> the demonstratives within the noun phrases,there is an asymmetry in distribution between the singular and plural demonstratives:the singular demonstrative can occur in either prenominal or postnominal positions(see (3)-(4)), while plural demonstratives are restricted to postnominal positions, asshown in (5)-(6). 32 This table is from Cheung (2003).3 I follow Cheung’s (2003) method in presenting Jingpo data. <strong>The</strong> data are presented in the followingfashion: in the first line, I shall present the Jingpo script followed by the IPA in the second line. <strong>The</strong>glossing is given in the third line, and finally, the English translation is in the fourth line. Also, notethat for sentence final particles which normally agree in number and person with the subject, I wouldsimply gloss them as ‘SFP-1Sg-DYM’, where ‘1Sg’ normally refers to the person and number features<strong>of</strong> the subject but not the object. DYM indicates dynamacy; whereas STA indicates stative.194


(3) Dai marau hpun grai tu tsom ai.tai ma au phun kai tu tsom ai that pine tree very grow beautiful SPF-3Sg-STA‘That pine tree grows up beautifully.’ (Dai and Xu, 1992: 368)(4) Marau hpun dai grai tu tsom ai.ma au phun tai kai tu tsom ai pine tree that very grow beautiful SPF-3Sg-STA‘That pine tree grows up beautifully.’ (Dai and Xu, 1992: 368)(5) *Dai-hte marau hpun grai tu tsom ai.tai the ma au phun kai tu tsom ai that-Pl pine tree very grow beautiful SPF-3Sg-STA(6) Marau hpun dai-hte grai tu tsom ai.ma au phun tai the kai tu tsom ai pine tree That-Pl very grow beautiful SPF-3Sg-STA‘Those pine tree grows up beautifully.’ (Dai and Xu, 1992: 368)<strong>The</strong>re are two word orders with singular demonstratives: either prenominal orpostnominal. As is observed in Cheung (2003), the same pattern follows when thedemonstrative co-occurs with other elements in the noun phrase, e.g. the noun,195


classifier and cardinal number: singular demonstratives can occur in the prenominaland postnominal position while plural demonstratives can only occur in thepostnominal position, as are exemplified in the following sentences: 4(7) Ndai n-gu kyin masum n ra nngai.n tai n ku kjin ma sum n a n ai this rice CL three NEG want SFP-1Sg(Lit.)‘I do not want three catties <strong>of</strong> this rice.’(8) *Ndai-hte n-gu kyin masum n ra nngai.n tai the n ku kjin ma sum n a n ai this-Pl rice CL three NEG want SFP-1SgCheung (2003) also observes that singular demonstratives can be insertedbetween the noun and the classifier, as is shown in (9). For plural demonstratives, theycannot occur between the noun and the classifier, because when they occur, cardinalnumbers cannot co-occur, as is shown in (12); therefore, the only possible position forplural demonstratives to co-occur with classifiers is the position after classifiers,shown in (10) and (11).4 Examples (7) - (12) are from Cheung (2003).196


(9) N-gu ndai kyin masum n ra nngai.n ku n tai kjin ma sum n a n ai rice this CL three NEG want SFP-1Sg(Lit.)‘I do not want three catties <strong>of</strong> this rice.’(10) *N-gu ndai-hte kyin n ra nngai.n ku n tai the kjin n a n ai rice this-Pl CL NEG want SFP-1Sg(11) N-gu kyin ndai-hte n ra nngai.n ku kjin n tai the n a n ai rice CL this-Pl NEG want SFP-1Sg‘I do not want these catties <strong>of</strong> rice.’(12) *N-gu ndai-hte kyin masum n ra nngai.n ku n tai the kjin ma sum n a n ai rice this-Pl CL three NEG want SFP-1Sg(12) is ungrammatical because the plural demonstrative indicate that the number<strong>of</strong> the entities denoted by the noun is more than one, but the value is unspecified.Since the number masum specifies the number <strong>of</strong> the entities, they two are197


incompatible with each other semantically. That is to say, ndai-hte cannot co-occurwith cardinal numbers.In Jingpo, demonstratives, either singular or plural, cannot occur between theclassifier and the cardinal number.(13) *N-gu kyin ndai masum n ra nngai.n ku kjin n tai ma sum n a n ai rice CL this three NEG want SFP-1Sg(14) *N-gu kyin ndai-hte masum n ra nngai.n ku kjin n tai the ma sum n a n ai rice CL this-Pl three NEG want SFP-1SgFinally, Jingpo demonstratives can appear after cardinal numbers. Since pluraldemonstratives are incompatible with cardinal numbers, only singular demonstrativeswill appear after cardinal numbers, shown in (15).(15) Sara MaLa a laika buk lahkong ndaisaa mala a laika puk lakho ntaiteacher Mala GEN book CL two this‘these two books <strong>of</strong> Teacher Mala’s’198


<strong>The</strong> distributional pattern <strong>of</strong> Jingpo demonstratives can be summarized as in (16):(16) Singular demonstratives can occur in the prenominal and postnominal position,while plural demonstratives normally occur in the postnominal position.Sometimes, plural demonstratives can also occur in the prenominal position.When plural demonstratives appear in the prenominal position, the pluraldemonstrative will reappear in the postnominal position, and there is a short pauseafter the prenominal plural demonstrative, as is shown in (17). But for singulardemonstrative there is no such short pause after the first instance <strong>of</strong> the demonstrative,as in (18). 5(17) Dai-hte, marau hpun dai-hte grai tu tsom ai.tai the ma au phun tai the kai tu tsom ai that-Pl pine tree that-Pl very grow beautiful SPF-3Sg-STA‘Those pine trees grow up beautifully.’(18) Dai marau hpun dai grai tu tsom ai.tai ma au phun tai kai tu tsom ai that pine tree that very grow beautiful SPF-3Sg-STA‘That pine trees grows up beautifully.’5 We will soon argue that in (18) the first instance <strong>of</strong> the singular demonstrative is actually the overtdefinite article in Jingpo, but at this stage we will not differentiate demonstratives from the definitearticle, and we will call both <strong>of</strong> them demonstratives.199


<strong>The</strong> complete distributional pattern <strong>of</strong> Jingpo demonstratives we have observedis shown in (19)-(21):(19) a. [Dem-Sg > N]b. [N >Dem-Sg]Singular demonstratives can be either prenominal or postnominal.c. [Dem-Sg > N> Dem-Sg]d. [Dem-Sg > N> Dem-Pl]Singular demonstratives can co-occur in the prenominal and postnominal postion.When the plural demonstrative is used postnominally, the singular demonstrative isable to occcur in the prenomianl postion.e. [N >Dem-Pl]f. *[Dem-Pl > N]g. [Dem-Pl , > N> Dem-Pl]Plural demonstratives normally do not occur in the prenominal postion. When a pluraldemonstrative appears in the prenominal position, the plural demonstrative has toappear in the postnominal position again, and there is an obligatory short pause afterthe prenominal plural demonstrative, as in (19g).(20) a. [N > Dem-Sg > Cl >Num]b. *[N >Dem-Pl >Cl >Num]c. *[N >Dem-Pl >Cl]d. [N > Cl > Dem-Pl]e. *[N > Cl > Dem-Sg >Num]200


f. * [N > Cl > Dem-Pl >Num]In Jingpo CL and Num cannot be separated by demonstratives. Pluraldemonstratives are incompatible with Num. Plural demonstratives can only occurafter Cl.(21) a. [N >Cl >Num > Dem-Sg]b. *[N >Cl >Num > Dem-Pl]c. [N >Cl > Dem-Pl]Demonstratives, either singluar or plural, can occur in the final position <strong>of</strong> thenoun phrase. <strong>The</strong> reason why plural demonstratives cannot follow Num is that theyare incompatible with each other semantically. <strong>The</strong>refore, plural demonstratives canoccur in the final position only if Num is absent.We propose that Jingpo prenominal demonstratives are D, whereas postnominaldemonstratives are adjective phrases as those in Mandarin and Yi.<strong>The</strong> phenomenon that a morpheme can be used both as demonstrative anddefinite article in a particular language is far from rare. Lyons (1999: 116) observesthat “in many languages the definite article is segmentally identical or very similar toone <strong>of</strong> the demonstratives (though differing in stress). An example <strong>of</strong> this is German,where unstressed der is the article, and the same form with stress is a demonstrative.In Danish, similarly, den is both a demonstrative and the free-form definite article.”In Mandarin, there is no definite article; therefore, demonstratives take up thetask <strong>of</strong> definiteness encoding. In Yi, there are both definite article and demonstratives.<strong>The</strong>refore, definiteness can be encoded by one <strong>of</strong> them, but Yi demonstratives and201


definite article cannot co-occur, just like the case in English.However, there are clear cases <strong>of</strong> demonstratives co-occurring with a lexicaldefinite article. Spanish and Catalan are good examples. When the article is present,the demonstrative must be in the postnominal position, a typical position foradjectives.(22) a. este pais (Spanish)this country‘this country’b. el pais estethe countrythis‘this country’(23) a. aquella ciutat (Catalan)thatcity‘that city’b. la ciutat aquellathe citythat‘that city’When demonstratives occur in adjective position, we can say that they have nodirect association with definiteness encoding; therefore a definite article is required toencode definiteness, as in Spanish and Catalan. <strong>The</strong> reason why demonstratives can202


only occur in definite expressions is that they are incompatible with indefiniteness.In Jingpo, like in German, singular demonstratives can occupy D position, andserve as definite articles. <strong>The</strong> definite articles can co-occur with demonstratives, justas the case in Spanish and Catalan. Based on the distribution <strong>of</strong> Jingpo demonstratives,we are led to a conclusion that Jingpo prenominal demonstratives, which arenecessarily singular, are definite articles.6.2 Adjectives in JingpoIn the previous section, we have examined the distribution <strong>of</strong> Jingpodemonstratives. In this section we will focus on the distribution <strong>of</strong> Jingpo adjectives.<strong>The</strong> typical structure <strong>of</strong> Jingpo noun phrases is exemplified by (24):(24) udi hkum sanitu 31 ti 31 khum 31 sa 31 nit 31egg CL seven‘seven eggs’<strong>The</strong> internal word order <strong>of</strong> the noun phrase structure in Jingpo strictly follows our<strong>Complex</strong> Classifier Hypothesis, repeated here as (25):(25)203


Similar to the demonstrative, the distribution <strong>of</strong> the adjective is relatively free inJingpo, as it may occur in either prenominal or postnominal positions, as illustrated inthe following examples:(26) gaba ai hpunka pa ai phun big RC-M tree‘big tree/tree’(27) hpun gabaphun treeka pa big‘big tree/trees’<strong>The</strong> pattern can be explained in the same manner we explain the distribution <strong>of</strong>Yi adjectives. (27) is the default word order, without any movement. We simplyassume that there is phonological null CL/Pr. <strong>The</strong> absence <strong>of</strong> the classifier may be dueto the fact that overt classifiers by default denote quantity. (27) does not denote thequantity. <strong>The</strong>refore, no overt classifier is allowed.(28)hpuntreegababig204


(26) involves incorporation and movement just as the case in Mandarin. <strong>The</strong>empty classifier will undergo incorporation to F and the complex head [CL+F] will bespelt out as ai – the traditionally called modifying marker (c.f. Dai and Xu, 1992).(29)ai hpun gabaRC-M treebigThis FP can occupy [Spec, CLP], and forms another predication relation withother predicates such as cardinal numerals. Gu and Dai (2002) has discussed the case<strong>of</strong> adjective conjunction in Jingpo. <strong>The</strong> data is in shown in (30).(30) hkye ai hte tsom ai nampan nhtan masumkhje 33 ai 33 the 31 tsom 33 ai 33 nam 31 pan 33 n 31 than 33 ma sum red RC-M and beautiful RC-M flower Cl three‘three bunches <strong>of</strong> red and beautiful flowers’ (Gu and Dai, 2002: 12)205


According to their analysis, (30) embodies the conjunction <strong>of</strong> two adjectivephrases before the head noun, and they argue that in terms <strong>of</strong> adjective conjunction,Jingpo is not different from Mandarin Chinese. <strong>The</strong>ir analysis has two flaws. On theone hand, they do not differentiate adjectives and [adjective+ai] in Jingpo; on theother hand, they do not explain why each adjective has to be followed by a RC-M aiin Jingpo, while in Mandarin Chinese the RC-M de will occur only once, i.e. after thelast adjective.We propose that this kind <strong>of</strong> complex conjunctional structure is not theconjunction <strong>of</strong> [hkye ai] and [tsom ai], as Gu and Dai (2001) suggests, but [hkye ainampan] and [tsom ai nampan]. <strong>The</strong> first NP is deleted due to phonological repetition.<strong>The</strong> structure <strong>of</strong> this phrase can be best illustrated by the following trees:(31)hhe nhtan masumand CL three206


ai nampan tsomRC-M tree beautifulai nampan hkyeRC-M treeredUnder our analysis, ai is the spell-out <strong>of</strong> a complex head [CL+F]. [hkye ai] and[tsom ai] are not constituents. <strong>The</strong>re is no way to conjoin them. <strong>The</strong> only possible wayis to conjoin the two FPs. (30) strongly supports our CL-as-Pr analysis in Jingpo.6.3 Further remarks on Jingpo demonstrativesWe have already proposed that Jingpo demonstratives have dual status. <strong>The</strong>y canbe <strong>of</strong> the category D and A at the same time. When they occur in the prenominalposition, they are D. When they occur in the postnominal position, they are AP.(32) a. Dai marau hpun grai tu tsom ai.tai ma au phun kai tu tsom ai that pine tree very grow beautiful SPF-3Sg-STA‘That pine tree grows up beautifully.’ (Dai and Xu, 1992: 368)207


. Marau hpun dai grai tu tsom ai.ma au phun tai kai tu tsom ai pine tree that very grow beautiful SPF-3Sg-STA‘That pine tree grows up beautifully.’ (Dai and Xu, 1992: 368)<strong>The</strong> different syntactic status <strong>of</strong> dai in (32a) and (32b) is show in (33a) and (33b)respectively. In (33a) dai occupies D. In (33b) dai is Dem.(33) a.daithatmarau hpunpine treeb.marau hpunpine treedaithat208


Since D has no ability to accommodate plurality, 6plural demonstratives cannever occur in D. <strong>The</strong>y can only serve as predicates, and they invariably occur in thepostnominal position. This prediction is evidenced by the following examples:(34) *Dai-hte marau hpun grai tu tsom ai.tai 33 -the 33 ma au phun kai tu tsom ai that-Pl pine tree very grow beautiful SPF-3Sg-STA(35) Marau hpun dai-hte grai tu tsom ai.ma au phun tai 33 -the 33 kai tu tsom ai pine tree that-Pl very grow beautiful SPF-3Sg-STA‘Those pine trees grow up beautifully.’ (Dai and Xu, 1992: 368)6.4 Jingpo possessivesIn this section, we will examine the derivational process <strong>of</strong> Jingpo possessives.<strong>The</strong> structure <strong>of</strong> typical Jingpo possessives is exemplified as follows:(36) MaLa a laikamala a laikaMala GEN book‘Mala’s book(s)’6 Contra Li’s (1999) proposal that the plural feature can be realized in D in languages like Mandarin,we believe that D has no ability to accommodate the plural feature.209


This structure is similar to Mandarin possessives, and can be illustrated by thefollowing tree:(37)a laika MaLaGEN bookMalaWhen demonstratives occur in the possessives, we can have the patternsexemplified in (38), (40), (42), and (43).(38) nye a laika ndai-nije a laika ntai nimy GEN book this-Pl‘these books <strong>of</strong> mine’<strong>The</strong> structure <strong>of</strong> this possessive phrase is illustrated by the following tree, inwhich the derivation <strong>of</strong> nye a laika (‘my book’) is identical to that <strong>of</strong> (37).210


(39)nye a aika ndai-nimy GEN book this-PlIn (39) the CL/Pr is not realized. It is empty. However, this CL can bephonetically realized. In (40) there is an overt classifier, and this classifier does notmove up to F.(40) nye a laika buk ndai -nije a laika puk ntai nimy GEN book CL this-Pl‘these books <strong>of</strong> mine’In this case, since CL does not move up to F, predicate inversion move is notpossible. Otherwise, the minimality condition will be violated. <strong>The</strong> syntactic structure<strong>of</strong> this possessive phrase is as follows:211


(41)nye a laika buk ndai –nimy GEN book CL this-PlIn the structure (41), if D is occupied by ndai, then we can have (42).(42) ndai nye a laika buk ndaini(hte)ntai je a laika puk ntai ni(the)the my GEN book CL this-Pl‘these books <strong>of</strong> mine’In this case, D is occupied by the definite article ndai. <strong>The</strong> other derivationalprocess is identical to (41).As mentioned before, a Jingpo plural demonstrative sometimes can appear in theprenominal position, but there should be a short pause after it, shown in (43).<strong>The</strong>refore, we would argue that the plural demonstrative is outside the possessiveDP. 77 <strong>The</strong> function <strong>of</strong> this prenominal plural demonstrative is unknown. It may function as the topic orsomething else. To determine its function requires more detailed research to the sentential structure,especially the CP system <strong>of</strong> the language, which is beyond the scope <strong>of</strong> this thesis and we will leave itfor future exploration.212


(43) ndai ni (hte), nye a laika buk ndaini(hte)ntai ni(the) je a laika puk ntai ni(the)this-Pl my GEN book CL this-Pl‘these books <strong>of</strong> mine’When cardinal numbers occur in the possessives, we can have the patternexemplified in (44), (46), (49), (51) and (53).(44) nye a laika buk ndai lahkongje a laika puk ntai lakhomy GEN book CL this two‘these two books <strong>of</strong> mine’This sentence is extremely tricky. According to my Jingpo native speakerconsultant, demonstratives are not allowed to occur between CL and Num. At the firstsight, (44) seems to violate this constraint. However, under our analysis, (44) isexactly what we predict to exist in Jingpo. In (44) the CL and the Num belong todifferent layers <strong>of</strong> predication relation. <strong>The</strong> derivational process <strong>of</strong> this possessive isillustrated as follows:213


(45)nye a laika buk ndai lahkongmy GEN book CL this twoIn (46), ndai occurs in the prenominal position; therefore, it is D. <strong>The</strong> wholederivational process involves two steps. <strong>The</strong> first step is to get the DP ndai laika:(46) nye a ndai laika buk lahkongje a ntai laika puk lakhomy GEN the book CL two‘these two books <strong>of</strong> mine’(47)<strong>The</strong>n this DP forms the possession relation with the possessor, generating thepossessive FP via predicate inversion. Finally the whole possessive phrase enters thepredication relation with NumP via CL.214


(48)buka ndai laika nye lahkongGEN ART book my twoIn the following example, again ndai occurs in the prenominal position, butdifferent from (47), it occurs before a possessive phrase rather than a bare nounphrase.(49) ndai nye a laika buk lahkongntai je a laika puk lakhothis my GEN book CL two‘these two books <strong>of</strong> mine’<strong>The</strong> derivational process <strong>of</strong> (49) can be illustrated in (50).215


(50)ndai nye a laika buk lahkongART my GEN book CL twoIn the following example, ndai occurs in the postnominal position. <strong>The</strong>refore, itis Dem rather than D.(51) nye a laika buk lahkong ndaije a laika puk lakho ntaimy GEN book CL two this‘these two books <strong>of</strong> mine’<strong>The</strong> derivational process <strong>of</strong> possessive is special in that the complement <strong>of</strong> CL inthis case is a complex predicate containing both NumP and DemP. We still stick toCinque’s remnant movement to derive the correct word order.216


(52)<strong>The</strong> following sentence is similar to (51). <strong>The</strong> only difference is that in this caseD is occupied by the definite article ndai.(53) ndai nye a laika buk lahkong ndaintai je a laika puk lakho ntaithis my GEN book CL two this‘these two books <strong>of</strong> mine’To sum up, Jingpo simple possessives only involve the possessor, the possessee,and the complex head [Pr+F] surfacing as a under our analysis. <strong>The</strong> derivationalprocess <strong>of</strong> getting this kind <strong>of</strong> simple possessives in Jingpo is identical to that inMandarin. <strong>The</strong> complex head [Pr+F] is the reflex <strong>of</strong> predicate inversion move in bothJingpo and Mandarin. What is unique about Jingpo is its complex possessives, that is,217


possessives with other predicates such as DemP and NumP. Since Jingpo has definitearticles in the prenominal position, it does not need to raise postnominal DemP to theprenominal position. <strong>The</strong>refore, in Mandarin, demonstratives will invariably occur inthe prenominal position by predicate inversion move; whereas in Jingpodemonstratives will not move. Rather Jingpo can directly merge its definite article inD. However, this operation is optional. Definite complex possessives may not containD. Rather they contain Dem. Unlike DemP in Mandarin, DemP in Jingpo does notmove to [Spec,FP]. It stays invariably in the postnominal position. 86.5 Jingpo relatives<strong>The</strong> structure <strong>of</strong> typical Jingpo relatives is exemplified as follows:(54)a. Mani ngai mari ai laika buk dai grai hti pyo ai.mani ai mai ai laikapuk tai kai thi pjo aiyesterday I buy RC-M book CL that very interesting SFP‘<strong>The</strong> book I bought yesterday is very interesting.’b. Dai Mani ngai mari ai laika buk grai hti pyo ai.tai mani ai mai ai laikapuk kai thi pjo aithat yesterday I buy RC-M book CL very interesting SFP‘<strong>The</strong> book I bought yesterday is very interesting.’8 It remains a question how DemP in Jingpo encodes definiteness.218


c. Mani ngai mari ai dai laika buk grai hti pyo ai.mani ai mai ai tai laika puk kai thi pjo aiyesterday I buy RC-M that book CL very interesting SFP‘<strong>The</strong> book I bought yesterday is very interesting.’In (54a), there is a postnominal Dem. In (54b) and (54c) there is a prenominal D.D is before RC in (54b); whereas D is before the relative head in (54c). <strong>The</strong> patternwe observe from (54) is summarized as follows:Table 2: Internal word order <strong>of</strong> the whole DP with RC in (54)Internal word order <strong>of</strong> the DP with RC in (54)54a RC RC-M N CL Dem54b D RC RC-M N CL54c RC RC-M D N CLD: dai (‘the’)N: laika (‘book’)RC-Marker: aiRC: Mani ngai mari (‘yesterday I bought e’)CL: bukDem: dai (‘that’)From example (54) and Table (2), we can observe that there are at least threedifferent structures for Jingpo relatives:(55 ) a. RC > RC-M > N > CL > (Num) > Demb. D > RC > RC-M > N > CL > (Num)c. RC > RC-M > D > N > CL > (Num)219


<strong>The</strong> derivational process <strong>of</strong> example (54a) is as follows: 9(56)<strong>The</strong> derivational process <strong>of</strong> (54b) is as follows: 10(57)9 Its constituency is shown in (55a).10 Its constituency is shown in (55b).220


<strong>The</strong> derivational process <strong>of</strong> (54c) is as follows: 11(58)From (56)-(58), we can see that Jingpo relatives, like Jingpo possessives, are derivedvia predicate inversion. <strong>The</strong> difference between relatives and possessives lies in thespell out <strong>of</strong> the complex head [Pr+F]. For possessives, the complex head is spelt outas a; whereas for relatives, it is spelt out as ai. Jingpo relatives are unique in that bothdemonstratives and definite articles can go with the relative head.Like Mandarin, Jingpo relatives can be stacked. For example:(59)a. Ngai mari ai grai tsom ai palong dai n mu mat sai.ai mai ai kai tsom ai palo tai n mu mat saiI buy RC very pretty RC dress that NEG see AUX SFP‘<strong>The</strong> pretty dress that I bought is lost.’11 Its constituency is shown in (55c).221


. Dai ngai mari ai grai tsom ai palong dai n mu mat sai.tai ai mai ai kai tsom ai palotai n mu mat saithat I buy RC very pretty RC dress that NEG see AUX SFP‘<strong>The</strong> pretty dress that I bought is lost.’c. Dai ngai mari ai grai tsom ai palong n mu mat sai.tai ai mai ai kai tsom ai palo n mu mat saithat I buy RC very pretty RC dress NEG see AUX SFP‘<strong>The</strong> pretty dress that I bought is lost.’Like Mandarin, the word order <strong>of</strong> the stacked RCs cannot be swapped. This isbecause the each <strong>of</strong> the two relative clauses observed in the above examples issemantically distinct:(60) a. [Ngai mari e ] ai palongI buy RC-M dressIn this case, the relative clause describes a property <strong>of</strong> the relative head. This propertyis not the inherent property <strong>of</strong> the relative head. We would call this kind <strong>of</strong>non-inherent-property denoting predicates stage-level predicates.b. [e grai tsom] ai palongvery pretty RC-M dressIn this case, the relative clause describes a property <strong>of</strong> the relative head. This property222


is the permanent property <strong>of</strong> the relative head. We would call this kind <strong>of</strong> permanent-property denoting predicates individual-level predicates.Relative clauses, which normally are stage-level predicates, are further awayfrom the relative head than individual-level predicate. <strong>The</strong> derivational process <strong>of</strong> thestacked relatives is as follows:(61)<strong>The</strong> reason why Jingpo and Mandarin relatives can be stacked is due to themechanism to derive relatives. Both Mandarin relatives and Jingpo relatives have thecomplex head, and relatives in these two languages are derived by the predicateinversion movement. If the relative clause is not derived by predicate inversion, then223


the stacking <strong>of</strong> relatives is barred. This exactly helps to explain the patterns weobserve in Yi. <strong>The</strong> reason why Yi cannot stack relatives is that Yi relatives are derivedby remnant movement rather than predicate inversion. Suppose we have the reducedrelative clause with classifier in Yi, as in (62).(62) co ssa kuo maꊿ ꌳ ꈄ ꂷperson brave CL‘a person who is brave/a brave person’If we want to stack a relative clause to this nominal, the only possibility is to usethe definite article su, because it is hard to determine which classifier should beemployed, since the specifier position is a DP rather than an NP, and we know that thechoice <strong>of</strong> classifier is completely determined by the NP at [Spec, CLP]. However inthis case, what is at [Spec, CLP] is not an NP any more. It is an FP. If we assume thatsu is used in this context, then by our analysis, sentence (63) will be readily available.However, (63) does not mean “the brave person that has come”. Rather it is a typicalpossessive construction in Yi, which means “the coming <strong>of</strong> a brave person/a braveperson’s coming.” <strong>The</strong>refore, both ways are ruled out. <strong>The</strong>re is no way for Yi relativesto stack.(63) co ssa kuo ma la o suꊿ ꌳ ꈄ ꂷ ꇁ ꀑ ꌠperson brave CL come ASP ART‘the coming <strong>of</strong> a brave person’224


(64)co ssa kuo maperson brave CLe la oe come ASP6.6 SummaryIn this chapter we have examined the complex nominal structure in Jingpo,which on the one hand shares properties with Mandarin and Yi, and on the other handdiffers significantly from Mandarin and Yi. For example, Mandarin has no definitearticle, whereas Jingpo and Yi have, but Jingpo differs from Yi in that it is possible forits demonstratives to co-occur with the definite article, which is banned in Yi.Mandarin and Jingpo have the derived complex head [Pr+F], which is spelt out as de225


and ai respectively, but Yi has no such derived complex head. Jingpo generatespossessives and relatives through predicate inversion; therefore in both Mandarin andJingpo, RCs can be stacked, and in both these two languages relatives are N-final. <strong>The</strong>following chart summarizes the differences between Mandarin, Jingpo and Yi.Table 3: Differences between Mandarin, Jingpo and Yi.Mandarin Jingpo YiArticle No Yes YesCo-occurrence <strong>of</strong> ART and Dem No Yes NoDerived <strong>Complex</strong> head Yes Yes NoMethod to generate RelativesPredicatePredicateRemnantinversionInversionMovementRelative Head Position N-final N-final N-initialStacking <strong>of</strong> RC Yes Yes No226


CHAPTER 7ConclusionsHaving worked through a range <strong>of</strong> phenomena relating to the interaction <strong>of</strong>classifiers, possessives and relatives in Yi, Chinese, and Jingpo in Chapters 2-6, in thisfinal chapter I would like to briefly summarize the main analyses and conclusionsproposed in this dissertation.<strong>The</strong> main goal <strong>of</strong> this dissertation is to explore the syntax <strong>of</strong> DP-internalpredication, based on which different ways <strong>of</strong> specificity and definiteness encodingemployed by different languages can be accounted for.I adopt Bowers’ (1991, 1993, 2001) predication theory and assume that thepredication relation holding between the subject and the predicate is established by anobligatory functional category Pr. 1A predicate is the syntactic constituent thatexpresses a property ascribed to the subject. In classifier languages, this functionalhead can be phonetically realized as CL. I have applied this CL-as-Pr analysis to suchcomplex noun phrases as possessive construction and relative construction in bothMandarin and Yi.<strong>The</strong> central topic <strong>of</strong> my thesis is the Pr function <strong>of</strong> classifier in possessives andrelatives, which is directly related to two empirical observations:1. <strong>The</strong>re are two ways to form relatives and possessives in Cantonese and Yi. <strong>The</strong>first method is to use CL. <strong>The</strong> second way is to use some other functional1 This functional category Pr proposed by Bowers, in essence, is similar to the functional head R(ELATOR)proposed in Den Dikken (2006). <strong>The</strong> difference is that Bowers argues that subject is always in the specifierposition <strong>of</strong> PrP and the predicate is in the complement position <strong>of</strong> PrP, but the syntax <strong>of</strong> predication proposed inDen Dikken (2006) claims that the predication relations, which are always hierarchically asymmetric, arefundamentally nondirectional.227


morpheme.2. In Mandarin and Cantonese, possessive morphology and relative morphology turnout to be identical, i.e., the possessive marker and relative marker are the same: deand ge respectively. What is strange is that in Yi they are deemed to be differentaccording to the existing literature.<strong>The</strong>se two observations are translated into my line <strong>of</strong> inquiry into the syntax <strong>of</strong>complex nominals:3. What is the function <strong>of</strong> CL in possessives and relatives?4. Why are the possessive marker and relative marker so different in Yi? Are theyreally different?I start my inquiry by introducing <strong>The</strong> <strong>Complex</strong> Classifier Hypothesis. FollowingChierchia (1998) and Aoun and Li (2003), this work assumes that bare noun phrasesin Sino-Tibetan classifier languages come out <strong>of</strong> the lexicon as fully referentialarguments. DemP, Cardinal NumeralP and AP first combine with CL to form afunction, which is applied to the NP at [Spec, CLP], resulting in a completeproposition within a nominal.<strong>Complex</strong> Classifier Hypothesis:CL and DemP/NumP/AP form a complex classifier, serving as the predicate <strong>of</strong> NP at[Spec,CLP]228


I have illustrated that Mandarin demonstratives are more like Spanishdemonstratives. <strong>The</strong>y are <strong>of</strong> adjectival type.(5)shu ben nabook CL thatshu ben nabook CL that<strong>The</strong> surface word order <strong>of</strong> this CLP is derived by predicate inversion: CL movesup to a higher head F, 2hence extends the domain and creates the equidistancebetween the [Spec, CLP] and [Spec, FP] with respect to DemP. <strong>The</strong>refore, DemP cancross NP and move higher to [Spec, FP].Chinese-type demonstratives (i.e. demonstratives which are APs rather than D),cardinal numbers and adjectives share a common property. <strong>The</strong>y are predicates. CL isthe functional head, which links the argument NP with the predicate. This is the majorsyntactic claim <strong>of</strong> the <strong>Complex</strong> Classifier Hypothesis (CCH).2 This F encodes the feature [+ POSS]. It projects to (N+I)P (Szabolcsi 1994). Since 那 本 书 (that CL book) doesnot encode the possession semantics, the value <strong>of</strong> the feature F is [-POSS]. Only [+POSS] F will form anaccordance relation with the upper head D, the [-POSS] head will neither adjoin nor incorporate into D. <strong>The</strong> reasonwhy DemP cannot be base-generated at [Spec, FP] is that DemP must form a unit with CL first, because there is apredication relation holding between DemP and NP, which has to be established by CL.229


We examine different uses <strong>of</strong> classifiers in extended projections <strong>of</strong> definite andindefinite nouns both in Mandarin and Yi.In the extended projection <strong>of</strong> indefinite nouns in Yi, CL adjoins to aphonologically null D, which can be viewed as the indefinite article <strong>of</strong> Yi. After CLhas incorporated into D, the remnant CLP moves to [Spec, DP].(6)In the extended projection <strong>of</strong> definite nouns in Yi, there are two different ways toencode definiteness in Yi extended projection <strong>of</strong> nouns. <strong>The</strong> first way is to resort tothe definite article su in Yi. <strong>The</strong> second way is to resort to demonstratives. When suappears in definite nominals, it will occupy the position D. CL adjoins to this D via F(the head <strong>of</strong> (N+I)P or POSS in Szabolcsi’s (1994) sense). <strong>The</strong>refore definitenominals are marked by [CL+F+su]. After CL has incorporated into D, the remnantCLP moves to [Spec, DP].When demonstratives appear in definite nominals, D has no phonetic realization.230


CL adjoins to F below this empty D. <strong>The</strong> co-indexation <strong>of</strong> DemP and NP at[Spec,CLP] renders the NP definite, and in turn makes the whole CLP definite. <strong>The</strong>definite CLP moves to [Spec,DP] and checks the strong D feature <strong>of</strong> the null D.<strong>The</strong>refore, we conclude that the definite nominal with a demonstrative is marked by[CL+F]. After CL has incorporated into F, the remnant CLP moves to [Spec, DP].(7)<strong>The</strong>se two options for encoding definiteness in Yi is in the same spirit <strong>of</strong> theDoubly Filled Comp Filter (Keyser 1975), which says that C must be null inembedded clauses if [Spec, CP] is not null. <strong>The</strong> Yi data show that D must be null if[Spec, DP] carries the feature [+D]. This can explain why demonstratives cannotco-occur with the definite article in Yi.In the extended projection <strong>of</strong> indefinite nouns in Mandarin, we assume that thereis CL-to-F movement, causing CL to play a role in determining the referentialproperty <strong>of</strong> the extended projection <strong>of</strong> nouns. In Mandarin, it is the complement <strong>of</strong>CLP, rather than CLP itself, that moves to [Spec, DP] via [Spec, FP].231


(8) a.zhuozitableyuanroundb.<strong>The</strong> extended projection <strong>of</strong> definite nominals in Mandarin assumes the wordorder <strong>of</strong> Dem>Num>CL>A>N, as is exemplified in the following tree:232


(9)yuan zhuoziround tablezhangCLnathatsanthreeWe propose an analysis <strong>of</strong> possessive structures in classifier languages. <strong>The</strong>thesis adopts the existential approach to possessives, since it captures the empiricalfacts observed in both Yi and Mandarin that possessive sentences are inherentlyconnected with existential constructions. We propose that the possessor in possessivesis not base-generated at [Spec,PrP]. It is base-generated as the complement <strong>of</strong> anabstract PP, which is the complement <strong>of</strong> Pr. <strong>The</strong> possessor or the locative phrase canundergo predicate inversion to reach [Spec, PrP].We also propose that there is a nominal INFL element F in the structure <strong>of</strong>complex nominals in strict parallelism with verbal INFL. <strong>The</strong> existence <strong>of</strong> this233


functional head F is argued to help encode definiteness. This means the encoding <strong>of</strong>definiteness is not solely by D. Instead, it is decided jointly by D and its complementvia agreement. I propose that, parallel to the CP structure, DP has the followinginternal structure:(10)<strong>The</strong> functional head F only contributes to the specificity <strong>of</strong> the whole nominal,since it is natural that possession relation normally presupposes the existence <strong>of</strong> thepossessee.<strong>The</strong> nature <strong>of</strong> Mandarin multi-functional morpheme de is also addressed inthis work. In the same spirit <strong>of</strong> Den Dikken and Singhapreecha (2004), we proposethat Chinese de, neither D nor C, is a complex functional head.de= Pr+FIn a possessive construction, F has the [+POSS] feature, thus inducing the possessionsemantics. With a (reduced) relative clause, F has the [-POSS] feature, thus unable to234


induce the possession semantics. However, in both cases, de is the spell out <strong>of</strong> thecomplex head [Pr+F].(11) a.shubookZhangsanZhangsanb.shubookyouquinterestingIn addition, [+POSS] F forms an accordance relation with the upper head, whichis D. In Mandarin it is hard to see this accordance relation, since there is no definitearticle in Mandarin. However, we can find evidence from Yi, which has the definitearticle:235


(12) mu gat ax yi ma suꃅꇢ ꀉ ꑳ ꂷ ꌠMuga child CL ART‘Muga’s that child’(13)<strong>The</strong> classifier ꂷ adjoins to F, and then [CL+F] further adjoins to D. Yi differsfrom Mandarin in that the complex head [Pr+F] will not be spelt out as a newmorpheme. In Yi, if Pr is overtly realized as CL, then the complex head [Pr+F] will berealized as CL, and if Pr is null phonetically, then the complex head [Pr+F] will not bespelt out. Rather it will invariably adjoin to the definite article su. <strong>The</strong> complex head[Pr+F] in Yi always maintains an agreement relation with the upper head D. If the236


whole nominal phrase is definite, then D will be realized as the definite article su, andthe complex head [Pr+F] will adjoin to the definite article. If the whole nominal isindefinite, then the [-def] D will be null. In this case, we can either assume that thecomplex head [Pr+F] adjoins to this null indefinite article, or assume that the complexhead [Pr+F] stays in situ. In either way, there is an interaction between [Pr+F] and D.We also examine relative clauses in Yi. Under our analysis, su is the definitearticle, which can mark relative clauses. Yi classifiers can also mark relative clauses.Like Cantonese, Yi has two types <strong>of</strong> relatives.(14) a. [Pr+F+su]-type relativesb. classifier-type relatives(15)OP i237


su is shown to be optional. When D is empty, classifier will be incorporated intothis D position and perform the task <strong>of</strong> forming relatives, and mark the relative headas indefinite. We know that su is not the spell-out <strong>of</strong> the complex functional head[Pr+F]. It is D. <strong>The</strong> actual complex functional head is null in Yi, and this null complexfunctional head, together with the definite article, can be used as both possessivemarker and relative marker. In this way we prove that the possessive morphology andrelative morphology in Yi are identical, just like many other classifier languages.We also extend our CL-as-Pr analysis to Jingpo complex nominals. We focus onthe distribution <strong>of</strong> Jingpo demonstratives. Our analysis shows that the free positioning<strong>of</strong> demonstratives within Jingpo nominal structure is due to the dual status <strong>of</strong>demonstratives. When they occur in the prenominal position, they are D. When theyoccur in the postnominal position, they are AP modifiers. Since D has no ability toencode plurality, plural demonstratives can never occur in the D. <strong>The</strong>y can only serveas AP modifiers. <strong>The</strong>refore, they will invariably occur at the postnominal position.Mandarin and Jingpo have the derived complex head [Pr+F], which has overtphonetic realization, but Yi has no such derived complex head. Jingpo and Mandaringenerate possessives and relatives through predicate inversion; therefore in bothlanguages, RCs can be stacked, and in both languages relatives are N-final.<strong>The</strong> empirical and theoretical significance <strong>of</strong> the research can be summarized asfollows: empirically, it can explain the interaction between classifiers, possessivemarkers and relative markers; theoretically, it can <strong>of</strong>fer evidence for the parallelismexisting between DP and CP.238


REFERENCESAbney, S. P. 1987. <strong>The</strong> English Noun Phrase and Its Sentential Aspect. Doctoraldissertation, MIT.Allan, K. 1977. Classifiers. Language 53, 284-310.Aoun, J. and A. Y.-H. Li. 2003. Essays on the representational and derivationalnature <strong>of</strong> grammar: the diversity <strong>of</strong> Wh-construction. Cambridge, Mass: MITPress.Au Yeung, B. 2005. An Interface Program for Parameterization <strong>of</strong> Classifiers inChinese. Doctoral dissertation, <strong>The</strong> Hong Kong University <strong>of</strong> Science andTechnology.Bowers, J. 1991. <strong>The</strong> syntax and semantics <strong>of</strong> nominals. In Cornell UniversityWorking Papers in Linguistics 10, 1-30. (Proceedings <strong>of</strong> the First Semantics andLinguistic <strong>The</strong>ory Conference.) Department <strong>of</strong> Modern Languages andLinguistics, Cornell University, Ithaca, NY.Bowers, J. 1993. <strong>The</strong> Syntax <strong>of</strong> Predication. Linguistic Inquiry 24, 591-656.Bowers, J. 2001. Predication. In M. Baltin and C. Collins (eds.), <strong>The</strong> Handbook <strong>of</strong>Contemporary Syntactic <strong>The</strong>ory. Oxford: Blackwell.Bowers, J. 2002. Transitivity. Linguistic Inquiry 33,183-224.Chao, Y.-R. 1968. A Grammar <strong>of</strong> Spoken Chinese. Berkeley and Los Angeles:University <strong>of</strong> California Press.Chen, K. and D. Wu. 1998. Yi Grammar. Beijing: Central University for Nationalities239


Press.Cheng, L. and R. Sybesma. 1999. Bare and Not-So-Bare Nouns and the <strong>Structure</strong> <strong>of</strong>NP. Linguistic Inquiry 30, 509-42.Cheung, C.-H. C. 2003. An Investigation <strong>of</strong> the Jingpo Nominal <strong>Structure</strong>. MPhil.<strong>The</strong>sis, <strong>The</strong> Chinese University <strong>of</strong> Hong Kong.Chierchia, G. 1985. Formal Semantics and the Grammar <strong>of</strong> Predication. LinguisticInquiry 16, 417-443.Chierchia, G. 1998. Reference to kinds across languages. Natural LanguageSemantics 6, 339-405.Chomsky, N. 1977. On Wh-movement. In P. Culicover, T. Wasow and A. Akmajian(eds.), Formal Syntax. New York: Academic Press.Chomsky, N. 1993. A Minimalist Program for Linguistic <strong>The</strong>ory. In K. Hale and S. J.Keyer (eds.), <strong>The</strong> View from Building 20: Essays in Linguistics in Honor <strong>of</strong>Sylvain Bromberger, 1-52. Cambridge, Mass.: MIT Press.Chomsky, N. 1994. Bare Phrase <strong>Structure</strong>. In G. Webelhuth (ed.), Government andBinding <strong>The</strong>ory and the Minimalist Program. Oxford: Blackwell.Chomsky, N. 1999. Derivation by Phase. MIT Occasional Papers in Linguistics 18.Cambridge, Mass.: MITWPL.Chomsky, N. 2000. Minimalist Inquiries: the Framework. In R. Martin, D. Michaelsand J. Uriagereka (eds.), Step by Step: Essays on Minimalist Syntax in Honor <strong>of</strong>Howard Lasnik, 89-155. Cambridge, Mass.: MIT Press.Chomsky, N. 2001. Beyond explanatory adequacy. MIT Occasional Papers in240


Linguistics 20. Cambridge, Mass.: MITWPL.Cinque, G. 1999. Adverbs and Functional Heads: A Cross-Linguistic Perspective.Oxford: OUPCinque, G. 2005. Deriving Greenberg’s Universal 20 and Its Exceptions. LinguisticInquiry 36, 315-32.Collins, C. 1997. Local economy. Cambridge, Mass.: MIT press.Dai, Q.-X. and X.-J. Xu. 1992. A Grammar <strong>of</strong> Kachin. Beijing: Central University forNationalities Press.den Dikken, M. 1998. Predicate inversion in DP. In A. Alexiadou and C. Wilder (eds.),Possessors, Predicates and movement in the determiner phrase, 177-214.Amsterdam/Philadelphia: John Benjamins.den Dikken, M. 2006. Relators and Linkers: <strong>The</strong> Syntax <strong>of</strong> Predication, PredicateInversion, and Copulas. Cambridge, Mass.: MIT press.den Dikken, M. and P. Singhapreecha. 2004. <strong>Complex</strong> noun phrases and linkers.Syntax 7, 1-54.Diesing, M. 1992. Indefinites. Cambridge, Mass.: MIT Press.Dixon, R.M.W. 1969. Relative Clauses and Possessive Phrases in Two Australianlanguages. Language 45, 35-44.Enç, M. 1991. <strong>The</strong> semantics <strong>of</strong> specificity. Linguistic Inquiry 22, 1-25.Freeze, R. 1992. Existentials and other locatives. Language 68, 553-595.Givon, T. 2001. Syntax. Amsterdam/Philadelphia: John Benjamins PublishingCompany.241


Grimshaw, J. 1990. Argument <strong>Structure</strong>. Cambridge, MA: MIT Press.Grimshaw, J., and R. A. Mester. 1988. Light verbs and theta-marking. LinguisticInquiry 19, 205-232.Gu, Y. 2005. Complementization and the Morphemes ai/sai/na in Jingpo. Paperpresented at New Inflections on Grammaticalization 3. Santiago de Compostela,Spain.Gu, Y. 2006. <strong>The</strong>ories <strong>of</strong> Tense and Aspect, and the Analysis <strong>of</strong> Chinese Syntax. Paperpresented at the Symposium <strong>of</strong> Contemporary Linguistic <strong>The</strong>ory and Studies <strong>of</strong>Chinese. Beijing.Gu, Y. and Q.-X. Dai. 2002. A Structural Analysis <strong>of</strong> the Jingpo [NP + A]Constructions. In D.-X. Dai and Y. Gu (eds.), Modern Linguistic <strong>The</strong>ories andStudies on Chinese Minority Languages, 50-69. Beijing: Nationality Press.Hale, K. and S.J. Keyser. 1993. On argument structure and the lexical expression <strong>of</strong>syntactic relations. In K. Hale and S.J. Keyser (eds.), <strong>The</strong> view from Building 20:Essays in linguistics in honor <strong>of</strong> Sylvain Bromberger. Cambridge, Mass.: MITPress.Heim, I. and A. Kratzer. 1998. Semantics in Generative Grammar. Oxford: Blackwell.Higginbotham, J. 1985. On Semantics. Linguistic Inquiry 16, 547-94.Higginbotham, J. 1987. Indefiniteness and Predication. In E. Reuland and A. Meulen(eds.), <strong>The</strong> Representation <strong>of</strong> (in)definiteness, 43-70. Cambridge, Mass.: MITPress.Hu, S.-H. 2004. Yi Structural Particles. Beijing: Central University for Nationalities242


Press.Huang, C.-T. J. 1982. Logical relations in Chinese and the theory <strong>of</strong> grammar.Doctoral dissertation, MIT.Huang, C.-T. J. 1987. Existential sentences in Chinese and (in)definiteness. In E.Reuland and A. Meulen (eds.), <strong>The</strong> representation <strong>of</strong> (in)definiteness, 226–253.Cambridge, Mass.: MIT Press.Kayne, R. 1994. <strong>The</strong> Antisymmetry <strong>of</strong> Syntax. Cambridge, Mass: MIT Press.Keyser, S. J. 1975. A Partial History <strong>of</strong> the Relative Clause in English. In J. Grimshaw(ed.), Papers in the History and <strong>Structure</strong> <strong>of</strong> English. University <strong>of</strong>Massachusetts Occasional Papers in Linguistics. Amherst, Mass.Larson, R. 1988. On the double object construction. Linguistic Inquiry 19, 335-392.Larson, R. 1991. <strong>The</strong> Projection <strong>of</strong> DP and DegP, to appear in R. Larson(forthcoming), Essays on Shell <strong>Structure</strong> Projection. Routledge, New York.Li, Y.-H. A. 1985. Abstract Case in Chinese. Doctoral dissertation, USC.Li, Y.-H. A. 1999. Plurality in a Classifier Language. Journal <strong>of</strong> East AsianLinguistics 8, 75-99.Litip, T. 2004. From phrase structure to the Minimalist Program: <strong>The</strong> syntacticstructure <strong>of</strong> Altaic languages. Beijing: Central University for Nationalities Press.Liu, H.-Y. and D. Wu. 2004. On the <strong>Structure</strong> <strong>of</strong> YI "N+Num+Cl+Su". In J. Li andS.-H. Hu (eds.), <strong>The</strong> Research on the Classifiers in Sino-Tibetan Languages,83-101. Beijing: Central University for Nationalities Press.Longobardi, G. 1994. Reference and Proper Names. Linguistic Inquiry 25, 609-666.243


Lyons, C. 1999. Definiteness. Cambridge: Cambridge University Press.Matis<strong>of</strong>f, J. A. 1972. Lahu nominalization, relativization, and genitivization. In J.Kimball (ed.), Syntax and Semantics, Volume I, 237-57. Studies in LanguageSeries. Seminar Press, New York.Matthews, S. and V. Yip. 1994. Cantonese: A Comprehensive Grammar. London:Routledge.Matthews, S. and V. Yip. 2001. <strong>The</strong> <strong>Structure</strong> and Stratification <strong>of</strong> Relative Clauses inContemporary Cantonese. In H. Chappell (ed.), Sinitic Grammar: Synchronicand Diachronic Perspectives, 266-281. Oxford University Press.May, R. 1985.Logical Form. Cambridge, Mass.: MIT Press.Milsark, G. 1974. Existential Sentences in English. Doctoral dissertation, MIT.Partee, B.H. 1987. Noun Phrase Interpretation and Type-Shifting Principles. In J.Groenendijk et al. (eds.), Studies in Discourse Representation <strong>The</strong>ory and the<strong>The</strong>ory <strong>of</strong> Generalized Quantifiers. Foris, Dordrecht.Ning, C.-Y. 1996. De as a functional head in Chinese. In B. Agbayani, K. Takeda andS.-W. Tang (eds.), UCI Working Papers in Linguistics 1, 63-79.Rizzi, L. 1990. Relativized Minimality. Cambridge, MA: MIT PressShen, Y., Y.-J. He and Y. Gu. 2001. Generative grammar theory and Chinesegrammar research. [shengcheng yufa lilun yu hanyu yufa yanjiu]. Harbin:Heilongjiang Education Press.Simpson, A. 2002. On the status <strong>of</strong> modifying de and the structure <strong>of</strong> the Chinese DP.In S.-W. Tang and C.-S. Liu (eds.), On the formal way to Chinese languages.244


CSLI.Stowell, T. 1981. Origins <strong>of</strong> Phrase <strong>Structure</strong>. MIT DissertationSzabolcsi, A. 1994. <strong>The</strong> noun phrase, in F. Kiefer & K. Kiss (eds), <strong>The</strong> syntacticstructure <strong>of</strong> Hungarian, 197-174. San Diego: Academic Press.Travis, L. 1984. Parameters and effects <strong>of</strong> word order variation. Doctoral dissertation,MIT.Tang, C.-C. J. 1990. Chinese phrase structure and the extended X’-theory. Doctoraldissertation, Cornell University, Ithaca, N.Y.Walters, S. and N. Atqi. 2006. Existential clauses in Nosu Yi texts. Linguistics <strong>of</strong> theTibeto-Burman Area 29, 107-127.Williams, E. 1980. Predication. Linguistic Inquiry 11, 203-238.Yang, J. 2005. <strong>The</strong> evolution mechanism <strong>of</strong> numeral classifier phrase from CN-type toNC-type in Tibeto-Burman languages. In J. Li and S. Hu (eds.), <strong>The</strong> Research onthe Classifiers in Sino-Tibetan Languages. Beijing: Central University forNationalities.Yap, F.-H., Y. Gu and H.-Y. Liu. 2005. From nominalizer to stance marker: a syntacticaccount <strong>of</strong> some grammaticalization/pragmaticization phenomena. Paperpresented at the conference From Ideational to Interpersonal: Perspectives fromGrammaticalization. Leuven.245

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!