University College London, UK, daphne.wang.19@ucl.ac.uk
(joint work with Mehrnoosh Sadrzadeh, Samson Abramsky, and Víctor H. Cervantes)
Motivated by ideas of Firth and Harris, computational linguists argue that if two words often occur in the same context, they have similar meanings. Despite the successes of this idea in disambiguation tasks in NLP [1], the systematic existence of ambiguity in natural language and its shades and nuances are notchallenges faced head on. Contextuality and its degrees are well studied topics in Quantum Mechanics. Here, the mathematical framework of Contextuality-by-Default (CbD) has become a useful tool when dealing with systems that are signalling and will also be convenient for natural language. Our line of research comes closest to the concept combination examples of [2]. However, as shown in [3], neither of the 23 examples of [2] are truly contextual. In our work, we did find combinations that are truly contextual for the first time. These are from a dataset of rank-2 cyclic (verb, noun) phrases constructed from [4,5] and the probabilities are tabulated from occurrences in corpora (BNC and uKWaC). More general types of systems were considered too, initially using the sheaf-theoretic approach to contextuality, but we showed that none of them are truly contextual. For our rank-2 cyclic dataset, we made use of the degrees of signalling and direct influence from M-Contextuality to find quantitative empirical evidence that the context affects ambiguous senses and meanings of nouns and verbs differently. So far contextuality has not been found in more general systems, although nothing seems to preclude the existence of contextual examples in certain types of models.
[1] H. Schutze. Automatic Word Sense Discrimination. Computational Linguistics, 24(1): 97–123, 1998.
[2] P. D. Bruza, K. Kitto, B. J. Ramm, L. Sitbon. A probabilistic framework for analysing the compositionality of conceptual combinations. Journal of Mathematical Psychology, 67: 26–38, 2015.
[3] E. Dzhafarov, R. Zhang, V. Cervantes, J. Kujala. On Contextuality in Behavioral Data. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 374, 08 2015.
[4] M Pickering, S. Frisson. Processing ambiguous verbs: Evidence from eye movements, Journal of experimental psychology. Learning, memory, and cognition, 27:556–73, 2001.
[5] M.K. Tanenhaus, J.M. Leiman, M.S. Seidenberg. Evidence for multiple stages in the processing of ambiguous words in syntactic contexts, Journal of Verbal Learning and Verbal Behavior, 18(4):427–440, 1979.