Semantic transparency: challenges for distributional semantics

Semantic transparency: challenges for distributional semantics Melanie J. Bell Anglia Ruskin University melanie.bell@anglia.ac.uk Martin Sch¨afer Friedrich-Schiller-Universit¨at Jena post@martinschaefer.info Abstract Using data from Reddy et al. (2011), we present a series of regression models of semantic trans- parency in compound nouns. The results indicate that the frequencies of the compound constituents, the semantic relation between the constituents, and metaphorical shift of a constituent or of the com- pound as a whole, all contribute to the overall perceived level of transparency. While not proposing an actual distributional model of transparency, we hypothesise that incorporating this information into such a model would improve its success and we suggest some ways this might be possible. 1 Introduction Recently, a number of studies in distributional semantics have addressed the semantics of NN and AN compounds and phrases. Under the heading of compositionality, they often discuss phenomena that in the psycholinguistic and morphological literature are cast as issues of semantic transparency. In this paper, we are not proposing an actual distributional model of semantic transparency, but rather making some linguistic observations which have consequences for distributional models that attempt to capture the phenomenon. In Section 2, we introduce the notion of semantic transparency and its relation to compositionality in distributional semantics, and in Section 3 we present our descriptive framework for the semantics of complex nominals. Sections 4 and 5 describe the method and results, respectively, of our empirical study based on the data from Reddy et al. (2011), which we recode and use to build four regression models with semantic transparency as the dependent variable. Finally, Section 6 discusses the implications of our results for distributional models. 2 Semantic transparency and compositionality The term ‘semantic transparency’ aims to capture the intuitive difference felt between compounds like hogwash, meaning ‘nonsense’, and a compound like milkman. In the literature, semantic transparency is defined in two main ways. One is the idea that it can be linked to meaning predictability. Plag (2003, 46) states that words are semantically transparent if “[...] their meaning is predictable on the basis of the word-formation rule according to which they have been formed.” According to this definition, hogwash is clearly not transparent. But the meaning of milkman also does not seem predictable. Assuming the standard dictionary definition man who delivers milk to people’s houses, are we to assume that there is a word formation rule of the kind x who delivers y to people’s houses? This kind of definition seems ex- cessively restrictive. The second kind of definition uses analysability rather than predictability. A classic example is Zwitserlood (1994, 344), who writes that “[t]he meaning of a fully transparent compound is synchronically related to the meaning of its composite words [...]”. In this sense, milkman clearly is transparent because any possible usage will allow linking the interpretation in some way to the meanings of the constituent parts. But here the problem seems to be that the definition is too wide. Even in cases like buttercup, the name for the small flower with …

Content truncated to 500 words