I am happy to report on two new HLP lab papers on implicit learning in language and beyond that recently were accepted for publication:
- Qian, T., Jaeger, T. F., and Aslin, R. 2012. Learning to Represent a Multi-Context Environment: More than Detecting Changes. Frontiers in Psychology 3.
- Fine, A. B. and Jaeger, T. F. in press. Evidence for implicit learning in adult language processing. Cognitive Science.
The first paper by Ting Qian is an opinion piece on learning and theories of learning in a world in which evidence is presented sequentially and where deviations from the expected always carry with them ambiguity about the cause of such deviation. So, how do learners figure out how to construct sufficiently adequate (i.e. good in coverage, though not necessarily accurate in terms of assumptions about the causes) causal theories of the world?
Although this is a very general question, I think that this piece contains some thoughts that will hopefully be inspiring for language researchers, too. In particular, it points to a problem that psycholinguistic research is just beginning to address: How the hack can we understand each other despite rampant variability both across and within speakers?
There is, of course, a good deal of work on this question in speech perception (incl. several beautiful papers by Kraljic and Samuel over the last five years and some really nice work on cross-speaker generalization when adapting to non-native speech, Bradlow and Bent, 2008). This work has looked at environments where listeners are expected to adapt (learn) because there is an priori reasons to expect that adaptation is necessary (e.g., because a new speaker is encountered). But, as Ting et al discuss, changes in the underlying causal structure of an environment are not always that apparent (e.g., pronunciations of a speaker can change because she is tired). This raises a whole set of questions that has so far received very little attention.
More generally, there is –to the best of my knowledge– relatively little attention has been paid to the development of theories that make clear quantifiable predictions about phonetic adaptation (for some recent developments, see Sonderegger and Yu, 2010; Kleinschmidt and Jaeger, 2011, 2012). In this context, it strikes me that it would be interesting to apply models developed for phonetic category acquisition to this type of data (e.g., Vallabha et al., 2007; McMurray et al., 2009; Toscano and McMurray, 2010).
There also is a striking lack of similar studies beyond speech perception, although this seems to be becoming a hot topic with several recent papers on syntactic expectation adaptation (Farmer et al., 2011; Fine and Jaeger, 2011; Kamide, 2012; Fine et al., submitted; see also Kaschak and Glenberg, 2004 for related results) that build on work in syntactic priming (Chang et al., 2006; Reitter et al., 2011) — in particular the idea that syntactic priming is a consequence of implicit learning (Bock and Griffin, 2000; Chang et al., 2000, 2006; Kaschak, 2007; Kaschak et al., 2012) and recent work on syntactic priming in comprehension (e.g., Arai et al., 2007; Thothahiri and Snedeker, 2008). This is also where the second paper listed above fits in (Fine and Jaeger, in press). Alex re-analyzed data from the eye-tracking experiment on syntactic priming in comprehension presented in Thothathiri and Snedeker (2008) to test whether the strength of syntactic priming (the amount of expectation shift as measured in changes in gaze patterns) is proportional to the prediction error experienced while processing the prime. This would be expected if syntactic priming is a consequence of implicit learning. Specifically, both the connectionist model proposed by Chang et al. (2006) and rational learning theories makes this prediction (see Jaeger and Snider, submitted). We find that, indeed, more surprising primes seem to prime more. This is just a small re-analysis and we hope that there will be more (and better) work that addresses this question.
Finally, it seems that adaptation and implicit learning is receiving more and more attention in other domains of psycholinguistics, too. There is a number of papers on similar questions in pragmatic processing (e.g., Grodner and Sedivy, 2011) and prosodic processing (Kurumada et al., 2012; for exciting related work on production, see, e.g., work in Julia Hirschberg’s lab).