Abstract

Given the, recent evidence for t)robabilist;ic lnechanisms in models of hmnan ambiguity resolution, this paper investigates the plausibility of exl)loiting current wide-coverage, 1)rob- abilistic parsing techniques to model hmnan linguistic pertbrmance. In particular, we investigate the tmrtbrlnance of standard stoolrestic parsers when they are revised to operate incrementally, and with reduced memory resources. We present techniques fi)r ranking and filtering malyses together with experilnen- tal results. Our results confirm that stochas- tic parsers which adhere to these 1)sychologically motivated constraints achieve good pertYrlmmce. Memory can be reduced down to l% (compared to exhausitve sero:oh) without re- ducing recall and precision. Additionally, these models exhibit subsl, antially fisi;cr l)erfi)rmance. Finally, we argue that this general resull, is likely to hold fi)r more sophisticated, and i)sycholinguistically plausible, probalfilistic parsing 1nod- els.

Citations