Corpus-based discovery of semantic intensity scales

Human Language Technologies: The 2015 Annual Conference of the North American Chapter of the ACL, pages 483–493, Denver, Colorado, May 31 – June 5, 2015. c⃝2015 Association for Computational Linguistics Corpus-based discovery of semantic intensity scales Chaitanya Shivade†, Marie-Catherine de Marneffe§, Eric Fosler-Lussier†, Albert M. Lai∗ †Department of Computer Science and Engineering, §Department of Linguistics, ∗Department of Biomedical Informatics, The Ohio State University, Columbus OH 43210, USA. shivade@cse.ohio-state.edu, mcdm@ling.ohio-state.edu fosler@cse.ohio-state.edu, albert.lai@osumc.edu Abstract Gradable terms such as brief, lengthy and ex- tended illustrate varying degrees of a scale and can therefore participate in comparative con- structs. Knowing the set of words that can be compared on the same scale and the associ- ated ordering between them (brief < lengthy < extended) is very useful for a variety of lexical semantic tasks. Current techniques to derive such an ordering rely on WordNet to determine which words belong on the same scale and are limited to adjectives. Here we describe an extension to recent work: we in- vestigate a fully automated pipeline to extract gradable terms from a corpus, group them into clusters reflecting the same scale and estab- lish an ordering among them. This method- ology reduces the amount of required hand- crafted knowledge, and can infer gradability of words independent of their part of speech. Our approach infers an ordering for adjec- tives with comparable performance to previ- ous work, but also for adverbs with an accu- racy of 71%. We find that the technique is use- ful for inferring such rankings among words across different domains, and present an ex- ample using biomedical text. 1 Introduction Gradability (Sapir, 1944) is a property of words that identifies different degrees of the quality the word denotes. For example, adjectives such as large, huge and gigantic present different degrees of size or vol- ume. Similarly, adverbs such as approximately, al- most and roughly present different degrees of how accurate a measurement is. Thus, one of the charac- teristics of gradable terms is that they participate in a scale and can be ordered along that scale: for ex- ample, good < great < excellent (Kennedy, 2007). Another characteristic is that gradable terms can ap- pear in comparative constructions, e.g., “A is larger than B”. Such comparative judgments are a psycho- logical process that precedes judgments of counting, e.g., “A is twice as large as B” (Sapir, 1944). Modern NLP systems face the challenge of inter- preting language as close to human perception as possible. Modeling gradable terms as well as their associated meaning and ordering is an important as- pect of this challenge. Such information can be very useful for a variety of inference tasks, such as senti- ment analysis (Pang and Lee, 2008) and textual in- ference (Dagan et al., 2006). However, current lex- ical resources, like WordNet (Fellbaum, 1998), lack annotations capturing the gradability of words. This weakens the notion of similarity: although words such as small and minuscule illustrate varying de- grees of size, they are listed as synonyms in Word- …

Content truncated to 500 words