21.04.2013 Views

Eckhard Bick - VISL

Eckhard Bick - VISL

Eckhard Bick - VISL

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

2.2.5.1), without ultimately losing the syntactic or semantic information 251 residing<br />

in traditional word class definitions (like the categories of demonstrative, possessive,<br />

indefinite, interrogative and relative pronouns, for what I have morphologically<br />

defined as determiners [DET, number/gender-inflecting] and independent specifiers<br />

[SPEC, non-inflecting]). Likewise, sense-distinctions alone were not regarded as a<br />

sufficient criterion for lexeme-distinctness, reserving, for instance, the distinction of<br />

the etymologically unrelated, but in Brazilian Portuguese homographic, ‘fato’ and<br />

‘fa(c)to’ for the semantic level - to be performed by sense tag disambiguation. This<br />

way, a purely morphological approach to lexeme identity was possible.<br />

One of the main objectives of this dissertation has been to show that the<br />

Constraint Grammar approach, which from the beginning has stressed the<br />

importance of a parsing lexicon and grown from morphology into syntax (Karlsson,<br />

1995, p.11), is ideally suited for such progression, not alone towards more “delicate”<br />

syntax, but also with respect to notational filtering (constituent or dependency trees)<br />

and, ultimately, semantics and semantics-based applications (like MT), - provided<br />

the lexicon is upgraded along the way. It appears justified to say that the progression<br />

on the syntactic level, involving subclause function, clause level dependency<br />

markers and constituent tree transformation, has yielded quantifiable results<br />

comparable to what has been achieved for the “benchmark” surface syntax of<br />

ENGCG (cp. chapters 3.9 and 8.1).<br />

Though here I have only sketched the outlines of a semantic CG-level (in<br />

chapters 5 and 6), a working system has been implemented for the entire lexicon,<br />

and disambiguation rules have been written for comprehensive valency instantiation,<br />

as well as selected areas of polysemy resolution, showing that - at least in principle -<br />

Constraint Grammar can be used to address parsing tasks at these levels. Valency<br />

instantiation, though it could be performed in many other ways, shows how the CG<br />

formalism can be made to handle what basically is a unification process, exploiting<br />

unambiguous syntactic information that has already been established (tagged).<br />

Unlike ordinary disambiguation rules, most of the valency unification rules are<br />

relatively simple, REMOVE rules often involving only the target valency tag and<br />

two unbounded context conditions, one left, one right, checking for the presence of a<br />

matching argument 252 . On the other hand, rules targeting semantic features or<br />

semantic prototype tags cannot be based on established syntactic tags alone, but need<br />

to take into account other (semantic) ambiguous information, resulting in a rule type<br />

more reminiscent of the morphological and syntactic levels. Finally, there is ongoing<br />

work involving a translation-equivalent Constraint Grammar for Portuguese-Danish<br />

machine translation, which is basically a context conditioned mapping grammar,<br />

refining and correcting the translations obtained by using morphological, syntactic,<br />

valency- and semantic tags as polysemy discriminators. And though such a statement<br />

should be worded carefully, nothing indicates insurmountable difficulties on the<br />

immediate Constraint Grammar application horizon ...<br />

251 This information, no not morphologically explicit, is retained by means of secondary and function tags, which can be<br />

exploited to recreate the traditional pronoun classes in a given application.<br />

252 For instance: REMOVE () (NOT *1 @

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

Saved successfully!

Ooh no, something went wrong!