Domain adaptation for parsing
Time: 03:00pm - 04:00pm
Place: Ballantine 015
Eric Baucom and Levi KingWe compare two different methods in domain adaptation applied to constituent parsing: parser combination and cotraining, each used to transfer information from the source domain of news to the target domain of natural dialogs, in a setting without annotated data. Both methods outperform the baselines and reach similar results. Parser combination profits most from the large amounts of training data combined with a robust probability model. Co-training, in contrast, relies on a small set of higher quality data.
In category: Computational linguistics
Interface delay and child language optionality
Time: 11:15am - 12:15pm
Place: Speech and Hearing Building, Room C141
John Grinstead (The Ohio State University)We claim that the root nonfinite verb phenomenon, or Optional Infinitive Stage, is a subcase of the larger phenomenon of Interface Delay, as a result of which distinct domains of cognition experience gradual development of their capacity to communicate with one another. Root nonfinite verbs, then, are the result of the delayed development of temporal anaphora – a construct that requires the interaction of syntax with the representation of interlocutors’ common ground perspectives. In the absence of adult-like interplay between syntax and common ground representations, children adopt a presupposition of familiarity. Delayed development of temporal anaphora is taken to be parallel to the delay in nominal anaphora manifested in the overuse of null subjects in child Spanish and the overuse of other definites, including overt tonic and clitic pronouns and definite articles, in an array of languages. The phenomenon in both typically-developing child Spanish and in the grammars of Spanish-speaking children with specific language impairment (SLI) is difficult, though not impossible, to detect in spontaneous production data because of the portmanteau nature of tense and agreement morphology, combined with the null subject nature of the language. These obstacles led early work to overestimate children’s competence in verb finiteness marking, but later work, especially using receptive tasks, has shown that child Spanish speakers produce, and accept as grammatical, an array of nonfinite forms including the bare stem (e.g. ‘canta’), the morphological infinitive (e.g. ‘cantar’) and the bare progressive participle (e.g. ‘cantando’). Internal validation of these findings comes from the fact that productive measures of nonfinite verb use correlate with receptive measures, downstream functions of tense marking (such as overt subject licensing) are significantly worse in Spanish-speaking children with SLI than in children without it and a discriminant function analysis based on elicited production measures of verb finiteness yields fair to good sensitivity and specificity (89%) for distinguishing Spanish-speaking children with SLI from those without. Further downstream evidence of optionality stemming indirectly from Interface Delay comes from optional child-particular grammatical constructions that depend on tense, including subject-auxiliary inversion and nominative Case marking in child English.
In category: Child language acquisition
Unaccusative verbs induce interpretation errors in L2 sentence processing
Time: 02:30pm - 04:00pm
Place: Ballantine Hall 205
Hyun-Kyoung Susan SeoTemporarily ambiguous sentences (1) with verbs alternating between transitive and intransitive semantics (2) induced readers to incorrectly answer yes at non-negligible rates when asked whether Anna dressed the baby (Christianson, Hollingworth, Halliwell & Ferreira, 2001; Ferreira, Bailey & Ferraro, 2002). Readers also robustly answered yes when asked whether the baby played in the crib. Thus, an initial misinterpretation of the baby as object of dressed lingered even after successful reanalysis as the main-clause subject. Such a thematic/argument-structure misinterpretation is unexpected with strictly intransitive verbs, whether unaccusative (e.g. arrive) or unergative (e.g. yawn) (Juffs, 2004; Staub, 2007). However, research also suggests a general late closure strategy might affect processing of intransitive verbs (van Gompel & Pickering, 2001; Juffs, 2004). (1) While Anna dressed the baby played in the crib. (2) Anna dressed. / Anna dressed the baby. An experiment investigated effects of a general transitivity phrase-structure assumption, possibly modulated by unaccusative/unergative verb semantics, with 18 experimental sentences such as (3) and (4). After reading 76 sentences at their own pace, forty L1-Korean learners of L2-English were prompted to answer queries such as Did the editor arrive or Did the editor postpone the meeting after reading (3) and Did the player yawn or Did the player kick the ball after reading (4), by pressing either the yes or no button. (3) As the journalist arrived the editor postponed the meeting.(4) As the goalkeeper yawned the player kicked the ball. In contrast to NSs, L2ers (incorrectly) answered Yes to Did the editor arrive (at 70%) after reading the sentence (3), although they (correctly) responded Yes to Did the editor postpone the meeting (at 100%). However, L2ers answered Yes to Did the player yawn (at 10%) after reading (4). A thematic misinterpretation modulated by unaccusative/unergative split is also reflected in a mild garden path effect visible in reading times (RTs), echoed in NSs' RTs. This suggests that in processing a general late-closure reflex in the phrase-structure domain is blocked by argument-structure/thematic lexical information. This blocking might be less robust for L2 learners since L2 lexical access is under-routinized (Favreau & Segalowitz 1983; Segalowitz & Segalowitz, 1993; Segalowitz, 2003). These L2 results affirm the independence of argument-structure processing from syntax, since information in the semantic component lingers even as initial parses are rejected (Dekydtspotter & Sprouse, 2001)..
In category: Second language acquisition
JEvents v3.0.9 Stable
Copyright © 2006-2013
Powered by Joomla!®
Copyright © 2008 The Trustees of Indiana University | Copyright Complaints