Lila R. Gleitman, Felicia Hurewitz,1 Jesse Snedeker 2Kirsten Thorpe3 & John C. Trueswell1
1 University of Pennsylvania, 2 Harvard University, 3 Stanford University
gleitman@cattell.psych.upenn.edu
The development of on-line sentence processing abilities in children has received renewed interest with the advent of eyegaze listening techniques that help map out ongoing parsing decisions of children as they hear utterances that refer to the surrounding visual environment (Trueswell, Sekerina, Hill & Logrip, 1999). In response to instructions containing a temporary prepositional-phrase ambiguity (e.g., "Put the frog on the napkin into the box"), five-year-olds' initial eyegaze patterns and ultimate actions reflect a strong preference to link "on the napkin" to the verb as a destination argument, despite the presence of referential-scene information supporting a modifier analysis (two frogs, one on a napkin). Older children and adults readily use this referential-scene information to guide parsing commitments.
In this talk, we provide a synthesizing account of this study and a series of follow-up studies from our lab, all of which point toward a probabilistic multiple-constraint theory of parsing development. Under this account, adult and child sentence comprehension is inherently a 'perceptual guessing game', in which multiple probabilistic cues are used to recover detailed linguistic structure. These cues (e.g., lexical-distribution evidence, verb semantic biases, referential scene information) come 'on-line' (become automatized) gradually over the course of development, with more reliable and easier-to-track evidential cues to structure developmentally preceding less reliable and more-difficult-to-track cues. Also, children's difficulty with revising initial commitments is attributed to general information processing control differences between children and adults, perhaps linked to developing mechanisms for inhibition (e.g., Diamond & Taylor, 1996).
Under this account, younger children's syntactic ambiguity resolution abilities should disproportionately rely on lexico-syntactic cues because of their high reliability in the input and their tight coupling to event meaning (e.g., Fisher, Gleitman & Gleitman, 1991). For instance, children's strong preference to link "on the napkin" to the verb in the Trueswell et al. study is predicted to be due to overwhelming distributional and verb-semantic biases of 'put' to have a destination phrase. This verb-bias account was confirmed in a new eyegaze study (Snedeker, Thorpe & Trueswell, in preparation) that fully crossed verb preference and referential manipulations within a single experiment design, using the instrument/modifier ambiguity ("Tap the horse with the feather"). Results indicated that adults combine both verb bias and referential information to determine syntactic choice, whereas children rely exclusively on lexical preferences, tightly matching three independently assessed levels of verb-instrument-role bias.
Evidence will also be discussed showing that (a) children's failure to use referential scene information is not due to a lack of knowledge about referential specificity or an inability disambiguate spatially; and (b) reliance on verb preferences is not due to a bottom-up architectural bias (i.e., modularity) but rather cue reliability, since a highly reliable discourse trigger for NP-modifying structures (Which-questions) also contributes to child parsing preferences (Hurewitz et al., 2000, 2001). Finally, given that child eyegaze patterns from our studies suggest there are deficiencies in inhibiting initial referential and syntactic analyses, we suggest that individual adult differences in inhibition control should be examined, especially in low-verbal-working memory subjects, as a source of difficulty in ambiguity resolution, rather than differences in working memory capacity.
References
Trueswell, J. C., Sekerina, I., Hill, N. M., & Logrip, M. L. (1999). Cognition, 73, 89-134.
Diamond, A., & Taylor, C. (1996). Developmental Psychobiology, 29, 315-334.
Fisher, C., Gleitman, H., & Gleitman, L. R. (1991). Cognitive Psychology, 23, 331-392.
Snedeker, J., Thorpe, K. & Trueswell, J.C. (2001). Manuscript in preparation.
Hurewitz, F., Brown-Schmidt, S., Thorpe, K., Gleitman, L. R.& Trueswell, J. C. (2000). Journal of Psycholinguistic Research, 29, 597-626.
Hurewitz, F., Trueswell, J.C., Gleitman, L. R., & Brown-Schmidt, S. (2001). Presented at the 14th Annual CUNY Conference on Human Sentence Processing.