Florian Jaeger (Rochester)
Wed 13 Jan 2016, 11:00 - 12:30
Room G32, 7 George Square

If you have a question about this talk, please contact: Jon Carr (jcarr3)

Florian Jaeger (work with Dan Gildea, Masha Fedzechkina, Lissa Newport, and John Trueswell)

Functional biases have been hypothesized to affect language change and explain typological patterns. I’ll focus on two specific pressures on language processing and production that are well-established. The first pressure relates to the fact that linguistic communication takes place in the presence of noise, so listeners need to infer intended message from noisy input —making less probable message harder to infer (e.g., Levy, 2008; Norris & McQueen, 2008; Bicknell & Levy, 2012; Gibson et al., 2013; Kleinschmidt & Jaeger, 2015). The second pressure relates to memory demands during language processing, where longer dependencies are associated with slower processing (Gibson, 1998, 2000; Lewis et al., 2006; Vasishth & Lewis, 2005). Both pressures are well-known to affect language processing, including evidence from both experimental data (e.g., McDonald & Shillcock, 2003; Grodner & Gibson, 2005) and broad-coverage corpus studies (e.g., Demberg & Keller, 2008; Boston et al., 2010; Smith & Levy, 2013). This means that an ideal speaker (in the sense of ideal observers) should a) support low-probability —i.e., high information— messages with ‘better’ linguistic signals to the extent that this is warranted against the effort in implies (e.g., due to aiming for more precise articulations or due to articulation additional words, cf. Lindblom, 1990, Jaeger, 2006, 2013; Gibson et al., 2013) and b) aim for short dependencies (e.g., by reordering constituents, Hawkins, 2004, 2014).

Case study 1 asks whether actual natural languages have syntactic properties that increase processing efficiency, as would be expected if the processing efficiency biases language learning and/or change. Using data from five large syntactically annotated corpora, I show that natural languages have lower information density and shorter dependency lengths than expected by chance (Gildea & Jaeger, in prep; for dependency length, see also Gildea & Temperley, 2010). Previous work has found similar properties for phonological and lexical systems (e.g, Manin, 2006; Piantadosi et al., 2011, 2012; Wedel et al., 2013). The present work is the first to find that the same properties affect even the syntactic system (which involves considerably more complex latent structure and has often been assumed to be encapsulated from functional pressures).

Case studies 2 and 3 employ an miniature language learning approach to the same question. I show that learners of such languages restructure them in a way that improves both the inferability of messages and the dependency length (Fedzechkina et al., 2011, 2013, under review; Fedzechkina & Jaeger, 2015).Unlike approaches that rely on statistical modeling of typological data, miniature language learning does not suffer from data sparsity and can —if applied correctly— assess causality by directly manipulating the relevant factors. 

Some references to related work from my lab, available at https://rochester.academia.edu/tiflo/Papers

  • Fedzechkina, M., Chu, B., and Jaeger, T. F. submitted. ‘Long before short’ preference in a head-final artificial language: In support of dependency minimization accounts.
  • Fedzechkina, M., Newport, E., and Jaeger, T. F. accepted for publication. Balancing effort and information transmission during language acquisition: Evidence from word order and case-marking. Cognitive Science.
  • Fedzechkina, M., Jaeger, T. F., and Newport, E. 2012. Language learners restructure their input to facilitate efficient communication. Proceedings of the National Academy of Sciences 109(44), 17897-17902. [doi:10.1073/pnas.1215776109]
  • Gildea, D. and Jaeger, T. F. submitted. Language structure shaped by the brain: Human languages order information efficiently.