“Was It Good? It Was Provocative.” Learning the Meaning of Scalar Adjectives

Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics, pages 167–176, Uppsala, Sweden, 11-16 July 2010. c⃝2010 Association for Computational Linguistics “Was it good? It was provocative.” Learning the meaning of scalar adjectives Marie-Catherine de Marneffe, Christopher D. Manning and Christopher Potts Linguistics Department Stanford University Stanford, CA 94305 {mcdm,manning,cgpotts}@stanford.edu Abstract Texts and dialogues often express infor- mation indirectly. For instance, speak- ers’ answers to yes/no questions do not always straightforwardly convey a ‘yes’ or ‘no’ answer. The intended reply is clear in some cases (Was it good? It was great!) but uncertain in others (Was it acceptable? It was unprecedented.). In this paper, we present methods for inter- preting the answers to questions like these which involve scalar modifiers. We show how to ground scalar modifier meaning based on data collected from the Web. We learn scales between modifiers and infer the extent to which a given answer conveys ‘yes’ or ‘no’. To evaluate the methods, we collected examples of question–answer pairs involving scalar modifiers from CNN transcripts and the Dialog Act corpus and use response distributions from Mechani- cal Turk workers to assess the degree to which each answer conveys ‘yes’ or ‘no’. Our experimental results closely match the Turkers’ response data, demonstrating that meanings can be learned from Web data and that such meanings can drive prag- matic inference. 1 Introduction An important challenge for natural language pro- cessing is how to learn not only basic linguistic meanings but also how those meanings are system- atically enriched when expressed in context. For instance, answers to polar (yes/no) questions do not always explicitly contain a ‘yes’ or ‘no’, but rather give information that the hearer can use to infer such an answer in a context with some degree of certainty. Hockey et al. (1997) find that 27% of answers to polar questions do not contain a direct ‘yes’ or ‘no’ word, 44% of which they regard as failing to convey a clear ‘yes’ or ‘no’ response. In some cases, interpreting the answer is straightfor- ward (Was it bad? It was terrible.), but in others, what to infer from the answer is unclear (Was it good? It was provocative.). It is even common for the speaker to explicitly convey his own uncer- tainty with such answers. In this paper, we focus on the interpretation of answers to a particular class of polar ques- tions: ones in which the main predication in- volves a gradable modifier (e.g., highly unusual, not good, little) and the answer either involves an- other gradable modifier or a numerical expression (e.g., seven years old, twenty acres of land). Inter- preting such question–answer pairs requires deal- ing with modifier meanings, specifically, learning context-dependent scales of expressions (Horn, 1972; Fauconnier, 1975) that determine how, and to what extent, the answer as a whole resolves the issue raised by the question. We propose two methods for learning the knowledge necessary for interpreting indirect an- swers to questions involving gradable adjectives, depending on the type of …

Content truncated to 500 words