Reading behavior predicts syntactic categories

Proceedings of the 19th Conference on Computational Language Learning, pages 345–349, Beijing, China, July 30-31, 2015. c⃝2015 Association for Computational Linguistics Reading behavior predicts syntactic categories Maria Barrett and Anders Søgaard University of Copenhagen Njalsgade 140 DK-2300 Copenhagen S {dtq912,soegaard}@hum.ku.dk Abstract It is well-known that readers are less likely to fixate their gaze on closed class syn- tactic categories such as prepositions and pronouns. This paper investigates to what extent the syntactic category of a word in context can be predicted from gaze fea- tures obtained using eye-tracking equip- ment. If syntax can be reliably predicted from eye movements of readers, it can speed up linguistic annotation substan- tially, since reading is considerably faster than doing linguistic annotation by hand. Our results show that gaze features do dis- criminate between most pairs of syntactic categories, and we show how we can use this to annotate words with part of speech across domains, when tag dictionaries en- able us to narrow down the set of potential categories. 1 Introduction Eye movements during reading is a well- established proxy for cognitive processing, and it is well-known that readers are more likely to fixate on words from open syntactic categories (verbs, nouns, adjectives) than on closed category items like prepositions and conjunctions (Rayner, 1998; Nilsson and Nivre, 2009). Generally, readers seem to be most likely to fixate and re-fixate on nouns (Furtner et al., 2009). If reading behavior is af- fected by syntactic category, maybe reading be- havior can, conversely, also tell us about the syn- tax of words in context. This paper investigates to what extent gaze data can be used to predict syntactic categories. We show that gaze data can effectively be used to dis- criminate between a wide range of part of speech (POS) pairs, and gaze data can therefore be used to significantly improve type-constrained POS tag- gers. This is potentially useful, since eye-tracking data becomes more and more readily available with the emergence of eye trackers in mainstream consumer products (San Agustin et al., 2010). With the development of robust eye-tracking in laptops, it is easy to imagine digital text providers storing gaze data, which could then be used to im- prove automated analysis of their publications. Contributions We are, to the best of our knowl- edge, the first to study reading behavior of syntac- tically annotated, natural text across domains, and how gaze correlates with a complete set of syntac- tic categories. We use logistic regression to show that gaze features discriminate between POS pairs, even across domains. We then show how gaze fea- tures can improve a cross-domain supervised POS tagger. We show that gaze-based predictions are robust, not only across domains, but also across subjects. 2 Experiment In our experiment, 10 subjects read syntactically annotated sentences from five domains. Data The data consists of 250 sentences: 50 sentences (min. 3 tokens, max. 120 characters), randomly sampled from each of five different, manually annotated corpora: Wall Street Jour- nal articles (WSJ), Wall Street Journal headlines (HDL), …

Content truncated to 500 words