Polysemy

Polysemy To be presented at TLTL lab meeting October 23. 

A combined word/character neural machine translation model can show, character by character, how a reader’s anticipation of what is coming changes. Booth (1978) (elaborated by Hinchcliffe & Frey, 2014) argues that this polysemy is the essential pleasure of the sonnets. At any moment in a reading—for example, the end of a line or phrase, an unexpected word—we hold open many meanings of the words. These meanings are expanded, closed off, reinforced, or inverted by subsequent words, so the meaning of any particular words has a temporal quality. 

Wortham (2001), drawing on Bahktin’s heteroglossia, calls this movement in meaning poetic structure. Wortham’s method of discourse analysis is interested in interactional positioning, the ways in which speakers are positioned with respect to roles within narrative. Speakers use poetic structure to reposition the meanings of what has already been said as well as the relationships of the participants to one another. 

Using a hybrid word/character NMT (Luong & Manning, 2016) model over the sonnets (perhaps trained also on a corpus of Elizabethan English), we can show the polysemy of the sonnets. Given a partial text of a sonnet, the model yields predictions for what is coming next. We can iterate over the whole sonnet, collecting predictions and watching as they change. Poetic structure will be particularly visible in words whose likelihood of appearing fluctuates dramatically. Take Stephen Booth’s definitive commentaries on sonnets could serve as an oracle against which to compare our results. 

We could then take a NMT model from today’s language, and use it to show which of these are likely to be missed by today’s readers! 

Bakhtin, M. M. (2010). The dialogic imagination: Four essays (Vol. 1). University of texas press. Booth, S. (1978). Shakespeare’s Sonnets: Ed. with Analyt. comm. Yale University Press. Hinchliffe, J., & Frey, S. (2014, October 9). Shakespeare’s genius is nonsense. In Nautilus (18). http://nautil.us/issue/18/genius/shakespeares-genius-is-nonsense Luong, M. T., & Manning, C. D. (2016). Achieving open vocabulary neural machine translation with hybrid word-character models. arXiv preprint arXiv:1604.00788. Wortham, S. E. F. (2001). Narratives in action: A strategy for research and analysis. Teachers College Press.