Distributional approaches to modeling meaning in use are getting more and more attention because they take some types of contextual clues into account while simultaneously lending themselves to computational implementations. In such approaches, the meaning of a word is represented by a vector consisting of the relative occurrence of this word with respect to other words within a certain distance of it. One advantage of this representation is the easy assessment of the similarity between words. If two words are similar, then their vectors will also be similar, which is to say that they occur in a given corpus with similar words at similar frequencies. However, such measures of word similarity do not transfer easily to notions of sentence similarity, especially when we take into account various kinds of implicatures. For example, is (1) more similar to (2) or (3), and in what contexts?

 

(1) John ate some cookies

(2) John ate all of the cookies

(3) John did not eat all of the cookies

 

Because sentence similarity measures have many applications like paraphrase detection and short answer tasks (Koleva et al., 2014), it is important to be able to fully measure the similarity between two sentences. In this presentation, I use scalar implicature as a lens through which to examine two compositional models: Mitchell & Lapata (2010) and Coecke et al. (2010). The former uses simple operations between vectors in its composition and the latter is able to derive sentence meaning while taking into account the structure of the composed sentence. After a brief review of the most prominent theories of scalar implicature (Chierchia et al. 2011, Geurts 2010), I will extend each of the computation theories to question of sentence similarity as it pertains to examples such as (1)-(3), highlighting both the pitfalls and possibilities for the implementation of further pragmatic phenomena into distributional theories.

 

References

Chierchia, G., Fox, D., & Spector, B. (2011). The Grammatical View of Scalar Implicatures and the Relationship between Semantics and Pragmatics. In P. Portner, C. Maienborn, & K. von Heusinger (Eds.), Handbook of Semantics (Mouton de, pp. 1–43).

Coecke, B., Sadrzadeh, M., & Clark, S. (2010). Mathematical Foundations for a Compositional Distributional Model of Meaning. Linguistic Analysis, 36(345).

Geurts, B. (2010). Quantity Implicatures. Cambridge University Press.

Koleva, N., Horbach, A., Palmer, A., & Ostermann, S. (2014). Paraphrase Detection for Short Answer Scoring. NEALT Proceedings Series, 22, 59–73.

Mitchell, J., & Lapata, M. (2010). Composition in Distributional Models of Semantics. Cognitive Science, 34(8), 1388–1429.