Identifying Semantic Relations and Functional Properties of Human Verb Associations

Proceedings of Human Language Technology Conference and Conference on Empirical Methods in Natural Language Processing (HLT/EMNLP), pages 612–619, Vancouver, October 2005. c⃝2005 Association for Computational Linguistics Identifying Semantic Relations and Functional Properties of Human Verb Associations Sabine Schulte im Walde and Alissa Melinger Computational Linguistics and Psycholinguistics Saarland University Saarbr¨ucken, Germany � schulte,melinger � @coli.uni-sb.de Abstract This paper uses human verb associations as the basis for an investigation of verb properties, focusing on semantic verb re- lations and prominent nominal features. First, the lexical semantic taxonymy Ger- maNet is checked on the types of clas- sic semantic relations in our data; verb- verb pairs not covered by GermaNet can help to detect missing links in the tax- onomy, and provide a useful basis for defining non-classical relations. Second, a statistical grammar is used for determin- ing the conceptual roles of the noun re- sponses. We present prominent syntax- semantic roles and evidence for the useful- ness of co-occurrence information in dis- tributional verb descriptions. 1 Introduction This paper presents an examination of a collection of semantic associates evoked by German verbs in a web experiment. We define semantic associates here as those concepts spontaneously called to mind by a stimulus word. In the current investigation, we assume that these evoked concepts reflect highly salient linguistic and conceptual features of the stim- ulus word. Given this assumption, identifying the types of information provided by speakers and dis- tinguishing and quantifying the relationships be- tween stimulus and response can serve a number of purposes for NLP applications. First, the notion of semantic verb relations is crucial for many NLP tasks and applications such as verb clustering (Pereira et al., 1993; Merlo and Steven- son, 2001; Lin, 1998; Schulte im Walde, 2003), the- saurus extraction (Lin, 1999; McCarthy et al., 2003), word sense discrimination (Sch¨utze, 1998), text in- dexing (Deerwester et al., 1990), and summarisation (Barzilay et al., 2002). Different applications in- corporate different semantic verb relations, varying with respect to their demands. To date, limited effort has been spent on specifying the range of verb-verb relations. Morris and Hirst (2004) perform a study on lexical semantic relations which ensure text co- hesion. Their relations are not specific to verb-verb pairs, but include e.g. descriptive noun-adjective pairs (such as professors/brilliant), or stereotypical relations (such as homeless/drunk). Chklovski and Pantel (2004) address the automatic acquisition of verb-verb pairs and their relations from the web. They define syntagmatic patterns to cover strength, enablement and temporal relations in addition to synonymy and antonymy, but they do not perform an exhaustive study. We suggest that an analysis of hu- man verb-verb associations may identify the range of semantic relations which are crucial in NLP ap- plications. We present a preparatory study where the lexical semantic taxonymy GermaNet (Kunze, 2000; Kunze, 2004) is checked on the types of classical se- mantic verb relations1 in our data; verb-verb pairs not covered by GermaNet can help to detect missing links in the taxonomy, and provide an empirical ba- sis for …

Content truncated to 500 words