We recently submitted a research review on “Speech perception and generalization across talkers and accents“, which provides an overview of the critical concepts and debates in this domain of research. This manuscript is still under review, but we wanted to share the current version. Of couse, feedback is always welcome.
In this paper, we review the mixture of processes that enable robust understanding of speech across talkers despite the lack of invariance. These processes include (i) automatic pre-speech adjustments of the distribution of energy over acoustic frequencies (normalization); (ii) sensitivity to category-relevant acoustic cues that are invariant across talkers (acoustic invariance); (iii) sensitivity to articulatory/gestural cues, which can be perceived directly (audio-visual integration) or recovered from the acoustic signal (articulatory recovery); (iv) implicit statistical learning of talker-specific properties (adaptation, perceptual recalibration); and (v) the use of past experiences (e.g., specific exemplars) and structured knowledge about pronunciation variation (e.g., patterns of variation that exist across talkers with the same accent) to guide speech perception (exemplar-based recognition, generalization).
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 Read the rest of this entry »
We hope to see y’all at CUNY in a few weeks. In the interest of hopefully luring to some of our posters, here’s an overview of the work we’ll be presenting. In particular, we invite our reviewers, who so boldly claimed (but did not provide references for the) triviality of our work ;), to visit our posters and help us mere mortals understand.
- Articulation and hyper-articulation
- Unsupervised and supervised learning during speech perception
- Syntactic priming and implicit learning during sentence comprehension
- Uncovering the biases underlying language production through artificial language learning
Interested in more details? Read on. And, as always, I welcome feedback. (to prevent spam, first time posters are moderated; after that your posts will always directly show)
In a recent PLoS one article, Healey, Purver, and Howes (2014) investigate syntactic priming in conversational speech, both within speakers and across speakers. Healey and colleagues follow Reitter et al (2006) in taking a broad-coverage approach to the corpus-based study of priming. Rather than to focus on one or a few specific structures, Healey and colleagues assess lexical and structural similarity within and across speakers. The paper concludes with the interesting claim that there is no evidence for syntactic priming within speaker and that alignment across speakers is actually less than expected by chance once lexical overlap is controlled for. Given more than 30 years of research on syntactic priming, this is a rather interesting claim. As some folks have Twitter-bugged me (much appreciated!), I wanted to summarize some quick thoughts here. Apologies in advance for the somewhat HLP-lab centric view. If you know of additional studies that seem relevant, please join the discussion and post. Of course, Healey and colleagues are more than welcome to respond and correct me, too.
First, the claim by Healey and colleagues that “previous work has not tested for general syntactic repetition effects in ordinary conversation independently of lexical repetition” (Healey et al 2014, abstract) isn’t quite accurate.
And it’s that time of the year again. Time to take stock. This last year has seen an unusual amount of coming and going. It’s been great to have so many interesting folks visit or spend time in the lab.
- Masha Fedzechkina defended her thesis, investigation what artificial language learning can tell us about the source of (some) language universals. She started her post-doc at UPenn, where she’s working with John Trueswell and Leila Gleitman. See this earlier post.
- Ting Qian successfully defended his thesis on learning in a (subjectively) non-stationary world (primarily advised by Dick Aslin and including some work joint with me). His thesis contained such delicious and ingenious contraptions as the Hibachi Grill Process, a generalization of the Chinese Restaurant Process, based on the insight that the order of stimuli often contains information about the structure of the world so that a rational observer should take this information into account (unlike basically all standard Bayesian models of learning). Check out his site for links to papers under review. Ting’s off to start his post-doc with Joe Austerweil at Brown University.
Post-docs Read the rest of this entry »
Speech recognition: Recognizing the familiar, generalizing to the similar, and adapting to the novel
At long last, we have finished a substantial revision of Dave Kleinschmidt‘s opus “Robust speech perception: Recognize the familiar, generalize to the similar, and adapt to the novel“. It’s still under review, but we’re excited about it and wanted to share what we have right now.
The paper builds on a large body of research in speech perception and adaptation, as well as distributional learning in other domains to develop a normative framework of how we manage to understand each other despite the infamous lack of invariance. At the core of the proposal stands the (old, but often under-appreciated) idea that variability in the speech signal is often structured (i.e., conditioned on other variables in the world) and that an ideal observer should take advantage of that structure. This makes speech perception a problem of inference under uncertainty at multiple different levels Read the rest of this entry »
(This is another guest post by Klinton Bicknell.)
This is an update to my previous blog post, in which I observed that post-version-1.0 versions of the lme4 package yielded worse model fits than old pre-version-1.0 versions for typical psycholinguistic datasets, and I gave instructions for installing the legacy lme4.0 package. As I mentioned there, however, lme4 is under active development, the short version of this update post is to say that it seems that the latest versions of the post-version-1.0 lme4 now yield models that are just as good, and often better than lme4.0! This seems to be due to the use of a new optimizer, better convergence checking, and probably other things too. Thus, installing lme4.0 now seems only useful in special situations involving old code that expects the internals of the models to look a certain way. Life is once again easier thanks to the furious work of the lme4 development team!
[update: Since lme4 1.1-7 binaries are now on CRAN, this paragraph is obsolete.]
One minor (short-lived) snag is that the current version of lme4 on CRAN (1.1-6) is overzealous in displaying convergence warnings, and displays them inappropriately in many cases where models have in fact converged properly. This will be fixed in 1.1-7 (more info here). To avoid them for now, the easiest thing to do is probably to install the current development version of lme4 1.1-7 from github like so:
Read on if you want to hear more details about my comparisons of the versions.