Language comprehension is sensitive to changes in the reliability of lexical cues.

摘要: Language comprehension is sensitive to changes in the reliability of lexical cues Alex B. Fine (afine@bcs.rochester.edu) Department Brain and Cognitive Sciences Linguistics, University Rochester Rochester, NY 14627 USA T. Florian Jaeger (fjaeger@bcs.rochester.edu) Computer Science, Abstract This paper tests hypothesis that language comprehen- ders update their beliefs about statistics throughout lifespan, this belief allows com- prehenders combine probabilistic linguistic according reliability. We conduct a multi-day sentence compre- hension study which cue syntactic structure manipulated between subjects. find as one de- creases, comprehenders come rely more on second structure. The results are consonant with ratio- nal models integration speech perception non- domains, thus suggesting unifying computational principle governing way humans use information across both perceptual higher-level cognitive tasks. Keywords: psycholinguistics; adaptation; process- ing; combination lawyer acknowledged judge had been unfair defendant. Probabilistic provide informa- tion can guide inferences during incremental processing, contributing processing efficiency (see Smith & Levy, 2008 for an explicit proposal along these lines). How- ever, relevant moving targets: such those mentioned above context- dependent validity (Bates MacWhinney, 1987) may vary depending speaker identity, context, dialect Tagliamonte, 2005 discussion variability syntax). Bates MacWhinney (1987) define product availability (how often present environment) leads correct inference, when present). How do cope maximize usefulness cues? current addresses ques- two-pronged hypothesis, framed spirit rational analysis (Anderson, 1990): Introduction In order process language, must make intended messages face uncertainty arising from noisy data ambiguity inherent lin- guistic signal. Research psycholinguistics suggests hu- mans accomplish task partially by capitalizing prob- abilistic well non-linguistic context (Jurafsky, 1996; Tanenhaus, Spivey-Knowlton, Eber- hard, Sedivy, 1995). For instance, sentences (1), temporarily interpretable direct object subject embedded complement. By-word reading times at point disambiguated (had been) correlated conditional probability structural representation as- signed parse given number probabilis- tic cues. One verb—the sen- tence complement occurring greater assert than ac- knowledge, based corpus norming (e.g., Trueswell, Kello, 1993; MacDonald, Pearlmut- ter, Seidenberg, 1994). A post-verbal noun phrase—if phrase unlikely be verb, it likely clause, thereby increasing continuation (Garnsey, Pearlmutter, Myers, Lotocky, 1997). Finally, if present, comple- mentizer . ) also serves strong struc- ture. Indeed, have shown all (1) (MacDonald et al., • A: Lifelong implicit learning: Throughout adulthood, continuously adjust estimates proba- bilistic comprehension. will refer adaptation (cf. Chang, Dell, Bock, 2006 references therein). B: Rational adaptation: Adaptation sense weight they assign particular cue. Preliminary evidence (A) comes prehension studies multiple levels (at phonetic level: Clayards, Aslin, Jacobs, 2008; Kraljic Samuel, 2007; Fine, Qian, Jaeger, 2010; Wells, Christiansen, Race, Acheson, 2009). (B) primarily (Clayards Kraljic, Brennan, 2008), though not necessarily terms hypotheses presented above. Of relevance Clayards al. (2008), who participants’ experience voice-onset time (VOT), category membership. participants group, distribution over VOT val- ues emerged course experiment low variance; other high variance. rationale manipulation