As requested by some, here are the slides from my 2015 CUNY Sentence Processing Conference plenary last week:
I’m posting them here for discussion purposes only. During the Q&A several interesting points were raised. For example –and I hope I recall this correctly (please correct me if I’m wrong):
- One question took as a starting point the apparent tension between the focus of my presentation (adaptation) and the well-established limitations to adults’ ability to learn new languages (apologies for not remember who asked this question). As I pointed out in my reply, there are indeed a series of interesting questions that center around this (apparent) conflict. I pointed to Pajak, Fine, Kleinschmidt, and Jaeger (under review) for a proposal that addresses this question. Specifically, rather than being in conflict with the view I was proposing, limitations of L2 learning and performance (such as continued interference from L1) follow naturally from the account I outlined. This account further correctly predicts that there’s also L2-to-L1 interference as well as interference from L_1 … L_k on L_(k+1) learning and use. This paper grew out of Bozena Pajak‘s thesis (UCSD) on multilingualism when we realized how similar that ideas she spelled out for multilingualism are to the ideas Dave and I develop for L1 speech processing.
- Peter Gordon (UNC)asked about–I think– the danger of providing a unifying adaptive account for phenomena that look similar only on the surface. I’d, of course, agree that there is such a danger, as with any theoretical explanation that seeks to explain, rather than to provide only a description of the data. As an example, Peter mentioned research on lexical decision where subjects are known to respond more slowly, for example, after seeing a phonologically well-formed nonce-word (i.e., a word that is harder to correctly classify as nonce word). This behavior might look adaptive, but in the literature on lexical decision (so Peter) it’s generally modeled as a shift in the decision criterion, rather than changes in distributions. I hope I didn’t get this wrong (if so, please correct me). My response to this would be that I indeed see no need to treat this example as being due to changes in distributions. In the view I was discussing one would attribute such a shift in the decision criterion to the utility function: when a participant realizes that there are items that are harder to classify, they increase the threshold confidence that is required before a button press is triggered. This would follow already under the assumption that any particular participant has some unknown utility function that trades off accuracy vs. speed in the lexical decision task (a rather uncontroversial assumption nowadays, I’d say; see e.g., Howes, Lewis, & Vera, 2009). In short, I fail to see the conflict between the criterion shift account and the (unifying) account in terms of utility.
At a more general level, I would like to add that I personally think that the field of psycholinguistics has certainly no shortage of highly specialized accounts that aim to explain a tiny portion of the available data. Quite to the contrary, we have a plethora of (named) accounts that describe one single aspect of language processing or even effects or paradigms. So, even if we were to find that a framework of the type I described (or the theories we have developed from it) is ultimately insufficient to explain certain specialized aspects of language processing (something that, I stress, remains to be shown), keeping in mind the bigger theoretical picture can proof tremendously productive in advancing our scientific understanding of how the brain works (and that, rather than understanding a paradigm, is presumably what we’re after). For example, simply stating that something is due to a criterion shift is only worth so much. Understanding why we shift the criterion (e.g., because there’s a goal of achieving a certain mean accuracy and that this can be captured by understanding the role of utility in decision making) holds the potential to understanding the deeper architectural principles of the brain/mind.
- Susanne Gahl (UC Berkeley) raised an objection to my claim that there was evidence that the targets of articulation are in perceptual space (I cited Guenther et al., 1998 among other work by, e.g., Perkell). Susanne pointed out that this is a debated notion and that other work suggests that targets are not exclusively (or even primarily?) in perceptual space. I’ll ask her to post some references here. There’s certainly evidence that adaptation as a response to perturbation is far from perfect: the corrections speakers make in response to perceived failures to hit the target they are aiming for are much smaller than the perturbation in auditory space. One reason for this might be that speakers also receive feedback about their own production through other sources (e.g., vibrations through the jaw). I see this as quite compatible with the claim I was making, but one can certainly interpret it differently.
There’s also evidence that suggests that production targets cannot be solely in perceptual space, but I hadn’t taken the experiments I am aware of as providing evidence against the claim that perceptual targets are important –or perhaps even primary–in planning and correcting articulation. Susanne pointed to bite block experiments. Rather than to put words into her mouth, I’ll let her speak to this (see comments below — you can subscribe the RSS feed of this page, if you want to get updates when comments are posted). In any case, I think that this type of literature is incredibly relevant to research on language production and yet often ignored in psycholinguistic theory formation.
- Jesse Snedecker (Harvard) asked whether what I presented was just a framework and whether it was compatible with “any theory”. I must admit I was a bit puzzled by this question, though I think it points to a very interesting discussion our field should have about what constitutes scientific progress. I’d be happy to discuss this here, if there’s interest, but the short answer is: yes, I was presenting a framework with the goal to guide future research. This framework makes clear predictions, and can certainly not be made to accommodate “any theory”. Additionally, I also pointed to (but did not discuss) the formal implementation and models we have for this framework (see Kleinschmidt and Jaeger, 2015, linked below, and, for syntactic adaptation e.g., Fine et al., 2010; Kleinschmidt et al., 2012). At a deeper level though I either misunderstood Jesse’s question or would take it to point to a rather fundamental disagreement we would have about the value of (good) frameworks, their role in scientific progress, as well as the typical strength of what we tend to call ‘theories’ in psycholinguistics (most of which I would prefer to refer to as “hypotheses” and which often lack in formal specification or even verbal precision, considerably limiting their predictive power).
There’s a good chance that misunderstood and (against my intentions) misrepresented some of the questions. If so, please let me know. You can email me and I will edit the questions or you can comment below to correct me. If you have never commented on this blog, the first comment requires confirmation from my side. All later comments are automatically posted.
For those interested in the ideas that are pointed to in this presentation, I’m attaching some papers from the lab that discuss them in more appropriate detail:
- Kleinschmidt and Jaeger (2015) – currently in press at Psychological Review: for a generative adaptive framework of speech perception and a review of the relevant literature on situation/talker/group-specificity.
- Fine, Jaeger, Farmer, and Qian (2013-PLoS One) relates these ideas to syntactic processing and reviews the literature available at that point (for a more in-depth investigation of lexical and syntactic adaptation during eye-tracking reading, see also Farmer et al., 2014 and Yan and Farmer’s presentation at this year’s CUNY).
- Buz, Tanenhaus, and Jaeger (under review) will be posted here at some point (ask me or Esteban, if you’re impatient). If you’re interested in the paradigm, please have a look at Esteban’s CogSci paper (Buz, Tanenhaus, and Jaeger, 2014).
- Kurumada and Jaeger (2015) – accepted at Journal of Memory and Language: elaborates ideas first sketched out in Jaeger and Ferreira (2013) about the role of goal-based adaptation in language production.