Measuring Thematic Fit with Distributional Feature Overlap

Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 648–658 Copenhagen, Denmark, September 7–11, 2017. c⃝2017 Association for Computational Linguistics Measuring Thematic Fit with Distributional Feature Overlap Enrico Santus1, Emmanuele Chersoni2, Alessandro Lenci3 and Philippe Blache2 enrico santus@sutd.edu.sg emmanuelechersoni@gmail.com alessandro.lenci@unipi.it philippe.blache@univ-amu.fr 1 Singapore University of Technology and Design 2 Aix-Marseille University 3 University of Pisa Abstract In this paper, we introduce a new distri- butional method for modeling predicate- argument thematic fit judgments. We use a syntax-based DSM to build a prototyp- ical representation of verb-specific roles: for every verb, we extract the most salient second order contexts for each of its roles (i.e. the most salient dimensions of typi- cal role fillers), and then we compute the- matic fit as a weighted overlap between the top features of candidate fillers and role prototypes. Our experiments show that our method consistently outperforms a baseline re-implementing a state-of-the- art system, and achieves better or compa- rable results to those reported in the liter- ature for the other unsupervised systems. Moreover, it provides an explicit represen- tation of the features characterizing verb- specific semantic roles. 1 Introduction Several psycholinguistic studies in the last two decades have brought extensive evidence that hu- mans activate a rich array of event knowledge dur- ing sentence processing: verbs (e.g. arrest) ac- tivate expectations about their typical arguments (e.g. cop, thief) (McRae et al., 1998; Altmann and Kamide, 1999; Ferretti et al., 2001; McRae et al., 2005; Hare et al., 2009; Matsuki et al., 2011), and nouns activate other nouns typically co-occurring in the same events (Kamide et al., 2003; Bick- nell et al., 2010). Subjects are able to determine the plausibility of a noun for a given argument role and quickly use this knowledge to anticipate upcoming linguistic input (McRae and Matsuki, 2009). This phenomenon is referred to in the lit- erature as thematic fit. Thematic fit estimation has been extensively used in sentence comprehen- sion studies on constraint-based models, mainly as a predictor variable allowing to disambiguate be- tween possible structural analyses.1 More in gen- eral, thematic fit is considered as a key factor in a variety of studies concerned with structural ambi- guity (Vandekerckhove et al., 2009). Starting from the work of Erk et al. (2010), several distributional semantic methods have been proposed to compute the extent to which nouns fulfill the requirements of verb-specific thematic roles, and their performances have been evaluated against human-generated judgments (Baroni and Lenci, 2010; Lenci, 2011; Sayeed and Demberg, 2014; Sayeed et al., 2015, 2016; Greenberg et al., 2015a,b). Most research on thematic fit estima- tion has focused on count-based vector representa- tions (as distinguished from prediction-based vec- tors).2 Indeed, in their comparison between high- dimensional explicit vectors and low-dimensional neural embeddings, Baroni et al. (2014) found that thematic fit estimation is the only benchmark on which prediction models are lagging behind state- of-the-art performance. This is consistent with Sayeed et al. (2016)’s observation that “thematic fit modeling is particularly sensitive to linguistic detail …

Content truncated to 500 words