Personalized grammar teaching on a tablet in inclusion classrooms

<strong>Personalized</strong> <strong>grammar</strong> <strong>teaching</strong> on a tablet in inclusion classroomsKarin Harbusch 1 , Christel-Joy Cameran 1 , Johannes Härtel 11 Department of Computer Science, University of Koblenz-Landau, Koblenz, Germany{harbusch|cameran|johanneshaertel}@uni-koblenz.de“Grammatical support tools allow people who have difficultieswith their <strong>grammar</strong> to make their computing and writing mucheasier. Good grammatical support can benefit even those withoutany learning disability or who do not think of themselves ashaving any significant <strong>grammar</strong> troubles.“ 1 Suitable for thisgoal, we present the personalized <strong>grammar</strong> <strong>teaching</strong> system IN-COMPASS for inclusion classrooms. The system can be accommodatedto different age groups.IN-COMPASS is a new surface system operating on top of thelanguage-technology based core system COMPASS (cf. [1] forits most recent visualization). COMPASS is an acronym forCOMbinatorial and Paraphrastic Assembly of Sentence Structure.In the core system COMPASS, the student constructs sentencesby composing syntactic trees out of lexically anchored“treelets” via a graphical drag&drop user interface (“scaffoldedwriting” [2]). In COMPASS, an advanced user can explore alllinguistic details in an interactive manner (self-paced and explorativelearning). A natural-language generator computes allpossible grammatically well-formed sentences entailed by thestudent-composed tree. It provides positive feedback if the student-composedtree belongs to the well-formed set, and negativefeedback otherwise. Proposals are based on a comparison betweenthe student-composed tree and its own trees (informativefeedback on demand). As feedback discusses the use of <strong>grammar</strong>rules, elementary syntactic knowledge is presupposed at thelearner’s side.In the following, we first focus on the basic features for <strong>grammar</strong><strong>teaching</strong> in the core system COMPASS. In turn, we outlinewhich details remain hidden by IN-COMPASS unless the pupilin an inclusive learning situation asks for more information.the learner first inflects a lemma according to the grammaticalfunction without deciding on its position in the constituentand/or in the sentence under construction, resp. At any time duringscaffolded writing, partial word ordering can be performed.These results (even if wrong) are kept during further expansionsof the dominance structure, e.g. by an Adjective.In an initial phase, the linguistic vocabulary used in visualizationsand feedback messages can be adjusted to the terminologyin line with the user’s preferences. As the system provides alreadylists of different terminologies in English and German, littlere-adjustment is necessary—same holds for the teachers in theinclusion classroom preparing the use of IN-COMPASS.In Fig. 1, we exemplify a potential student action which is notapproved by the generator. The disagreement is graphically realizedby not unifying the two treelets although the student hasmoved one on top of the other (as the arrow indicates). In theexample, we assume that the student had already successfullyconstructed a dominance tree for Anja baut ‘Anja builds’. In thistree, the student wants to add ihm ‘him’ (case DATIVE) at theDirect Object node. In Fig.1, it is still positioned close the desiredtarget leaf below the DOBJ node. However, it does not becomeattached because word form ihn (case ACCUSATIVE) isrequired instead. Feedback is supplemented by traffic-light inspiredcolor-coding. Red highlights forbidden actions.COMPASS provides intuitive graphical support for <strong>grammar</strong><strong>teaching</strong>. For instance, running on a tablet, the student moves(parts of) trees around with the finger or a stylus, resp. Especially,word-ordering—which is a difficult topic for learners ofGerman—is performed by the student as graphical dialog withthe system, i.e. by moving branches into field positions providedby the system (cf. the dark grey squared areas in the Figures).The syntactic <strong>grammar</strong> formalism underlying COMPASS, isPerformance Grammar [3]. Due to psycholinguistic reasons, itprovides a separation into dominance structures and linear ordering.“Scaffolded writing”—as performed in COMPASS—hasthe advantage that the students can concentrate on very smallproblems in L2 sentence production in isolation. For instance,1 Quoted from a description of the inclusive design research center atOCAD University, Canada (cf. http://idrc.ocad.ca/index.php/workshops-aamp-trainings/masters-program/13-resources/technical-glossary/37-grammatical-support-tools [last visit: 28th of June, 2013]).Figure 1. Illustration of a mismatch combining two treelets.In Fig. 2, the system’s warning reaction to ordering Anja bauteine Rakete ihm ‘Anja builds a rocket him’ is delineated. Thestudent has achieved this state by sorting all filled grammaticalfunction nodes in the grey squared area attached to the HeadVerb. In Fig. 1, this box is empty because so far, the student hasconcentrated on the dominance structure. Although word order iswrong, COMPASS does not reject any movement into fields butit indicates errors by inverting color to yellow so that the sen-

tence construction can go on while word ordering is suspendedfor the time being. In this manner, the system “teaches” evencomplicated cases such as Scrambling in German subordinateclauses to the student.Figure 2. Word order warning in Anja baut eine Rakete ihm.IN-COMPASS provides elementary dialogue features similar tothe core version COMPASS, i.e. the students can drag&dropavoiding typos on a tablet with the finger/stylo or on a PC usingthe mouse/touch pad, resp. As additional feature of IN-COMPASS, a read-aloud function is always available. It is indicatedby a loud-speaker button for every written text in order tosupport elementary school children in particular, but also dyslexicpupils in any classroom.Basically, a reduced lexical frame for an inflected word formexhibits its grammatical functions. The color-encoded fillers inthe core system —only providing distinguishable colors from thefeedback colors red/yellow/green — remain unchanged as thisalready allows for easily guiding the combining process withoutknowing any linguistic term. The naming of a grammatical functionis initially provided by the wh-question for the function(e.g., SUBJECT=WHO; cf. Fig. 3 for the lexical frame of baut‘builds’). Any Interrogative Pronoun in the light grey grammatical-functionlayer—same as any visualized element in (IN-)COMPASS—can be selected to elicit a feedback dialogue. Inreturn, the system provides more details on the function (e.g.,different filling options: for a subject, a Personal Pronoun or acomplex Noun Phrase). Most importantly, the system wonderswhether the learner knows already the—so far hidden—technicalterm by providing a list of grammatical functions to choosefrom. In case of a wrong answer, the system—as for any wronganswer—offers the correct answer on demand. Only if the userknew/wanted to know the term, the system suggests to replacethe wh-Pronoun by the linguistic term to be used by the system—asit is now learnt by the pupil. Thus, a personal level ofdetail becomes unveiled to the learner. As a consequence, a mixof Interrogative Pronouns and linguistic terms is displayed regardingto the user’s familiarity with <strong>grammar</strong> yet.Now we focus on the predefined layers of granularity hidingaway detailed linguistic information in IN-COMPASS—unlessthe user asks for. Thus, every pupil in the classroom can work inan individualized manner with the system following personalgoals (self-paced and explorative learning).Especially, users who are disabled or at a young age should notbe confronted with elaborate syntactic trees full of very abstractlinguistic terms they do not understand. Nevertheless, there hasto be an elementary agreement on the use of words in a sentencebetween user and system in order to give adequate feedback toany action by the learner (cf. discussions in the CALL literatureon parsing-based systems which often have difficulties to providehelpful comments to errors by the learner; see, e.g., [4]).Therefore, IN-COMPASS works with reduced student-composedtrees—like COMPASS—instead of sentences typed by the pupil.Thus, we completely avoid dealing with typos, missing words,hardly any punctuation although both systems, COMPASS andIN-COMPASS, resp., allow typing the desired word form. However,due to a nearest match procedure, only corrected lexicalframes can be selected/worked with.Figure 3. Reduced lexical frame for baut ‘builds’. Speakers supportstudents with reading problems.Moreover, the system allows training of word order variation,where the Subject has to be moved away from its often chosensentence-initial position. Of course, German speaking pupils arecapable of understanding any variation. However, teachers complainabout when grading pupils’ essays that the pupils do notexhibit this variation in writing very likely. Therefore, playingaround with sentences in order to understand the function of thesentence-initial position in different contexts can be elicited evenwith very reduced linguistic knowledge.Due to lack of space, here we can neither elaborate on more linguisticexploration modes (e.g., “<strong>teaching</strong>”/let pupils find outwhat it means to be Head of a phrase in order to understand whya Preposition is the Head of a PP by playing around with casevariation such as in auf den/dem Bürgersteig ‘onto/on the sidewalk’)nor discuss the appropriate accommodation to differentage groups in IN-COMPASS (e.g., the more but the shorter thefeedback by the system is the younger the pupils are in order toavoid overtaxing their attention span).Literature[1] Harbusch, K., Härtel, J. and Cameran, C.-J. “COMPASS III: Teaching L2 <strong>grammar</strong> graphically on a tablet computer”, in Proceedings of SLaTE –Speech & language technology in education, Grenoble, August 2013.[2] Harbusch, K. and Kempen, G. “Automatic online writing support for L2 learners of German through output monitoring by a natural-language paraphrasegenerator”, in M. Levy, F. Blin, C. Bardin Siskin and O. Takeuchi [Eds.], WORLDCALL, pp. 128-143, New York: Routledge, 2011.[3] Kempen, G. and Harbusch, K. “Performance Grammar: A declarative definition”, in M. Theune, A. Nijholt and H. Hondorp [Eds], ComputationalLinguistics in the Netherlands 2001, pp. 146–162. Amsterdam: Rodopi, 2002.[4] Heift, T. and Schulze, M. (Eds.) (2003). “Error diagnosis and error correction in CALL”, CALICO Journal, 20(3).