• LabmeetingSP15w14

Lab Meeting, Spring 2015, Week 14

Awe- and Aw-inspiring readings (or just stuff you think deserves a mention)

• Sjerps, M. J., & Reinisch, E. (2015). Divide and Conquer: How Perceptual Contrast Sensitivity and Perceptual Learning Cooperate in Reducing Input Variation in Speech Perception. Journal of Experimental Psychology: Human Perception and Performance. doi:10.1037/a0039028. Lexically guided phonetic recalibration (a la Norris et al. 2003). But carrier words are to make them more or less acoustically similar to the sound they usually end in. E.g., the parts of an /s/ word before the fricative are filtered to make them more /f/ and less /s/ like. This eliminates recalibration to an ambiguous [fs] sound. My (Dave's) reading: the added 'f'-ness from the filter explains the additional 'f'-ness of the ambiguous (but unfiltered) [fs] sound that normally is unexpected and leads to recalibration. (They interpret it in terms of perceptual normalization/lateral inhibition between previous time slices' cue values in TRACE).

• Morton, J. R., Sommers, M. S., & Lulich, S. M. (2015). The effect of exposure to a single vowel on talker normalization for vowels. The Journal of the Acoustical Society of America, 137(3), 1443–1451. doi:10.1121/1.4913456. Normally, when multiple talkers are jumbled together in an experiment it impairs comprehension (even for familiar talkers), relative to blocked talker presentation. This experiment shows that you can get rid of this multi-talker effect by playing a single /i/ vowel before each talker change. There's a substantially weaker effect of presenting the name of the talker instead of one of their vowels. They don't compare the effects of talker familiarity, which is too bad.

What we did over last week

Florian

1. Prepared talk for plenary at "Causality in the Language Science".

2. Revised design for the next need + participle study together with Scott F. We're trying to figure out what exactly is being learned when one sees a new unexpected structure. There are some links to Maryam's work, too, that we should chat about.
3. Continued working on paper with Dan Gildea about the processing efficiency of natural languages. A first draft is available upon request.

4. Tweeted a bit about nice papers I've read.

5. Watched Percy Liang's talk in a train to Leipzig. Pretty neat!
6. Met with an undergraduate who is potentially interested in building on Nikki Craycraft's work on syntactic adaptation.

7. Met with Maryam and Chigusa to talk about her planned Linguistic and Language Compass article.

8. Received an invitation to edit a special issue on phonological planning in Frontiers because of the awesome impact HLP lab articles had in Frontiers (apparently we are 96th percentile in terms of impact! Keep it up). Inclined to decline, unless one of you is interested in doing this and taking the lead?. Might be able to negotiate a free Frontiers article out of it.

Chigusa

Andrew

1. Found more HITs to include in the counts / demographic reports for various reports
2. Further cleanup of my demographic parsing script, because I found some nasty bugs

3. Ditched the RSS feed widget I was using on the new webpage and wrote my own using the Google Feed API to get the feed as JSON.

4. More work on the new webpage in general. Check out the draft version at https://www.hlp.rochester.edu/2015refresh/ If you have a good background picture to replace the neuron pic I grabbed from Google Image Search, then let me know.

5. Helped Chelsea fix the eyetracker set up in 415 so that Matlab, Eyelink, PyGaze, etc. experiments as well as ExBuilder can talk to the tracker host PC.

6. Created a Lab YouTube channel and uploaded the videos from our old webpage there instead of self hosting for the new webpage.

Olga

Esteban

Dave

1. Shot down the results I presented in lab meeting last week. They're entirely due to the collinearity in the design matrix (from the overlapping BOLD signals from nearby trials and fixed trial order across subjects). Moreover, the collinearity is a lot worse than I suspected (66% of trials' Beta values are correlated with another trial's Beta at >0.3).

2. Did some reading on whether or not talker familiarity makes it easier to switch out talker-specific representations. The literature looks like the consensus answer is no. This suggests to me that the story is a little more complicated than I thought when writing the "ideal adapter" paper.

3. Speaking of which: it's out!

4. Analyzed results of an additional condition in the shifted VOT distribution experiments I presented about a few weeks ago, where the mean /b/ VOT was -10ms and the mean /p/ was 30. I thought learning would be better in this condition than in the 20/60 condition, but it's surprisingly close. Or maybe no so surprisingly given that 30ms is very low for a mean /p/ VOT...
5. Cleaned up my MTurk results files from the very earliest experiment I ran so Andrew can more easily access the results.

1. (Dave, Last week)

1. Analyzed VOT fMRI data presented in lab meeting.
2. Gave a "big picture" talk in the CS department.
3. Ran an additional condition in teh VOT shift experiment (see above).

4. Hacked together a one-time qualification granting and HIT running script in Python because the MTurk Java command line tools have been silently nerfed and don't accept the "DoesNotExist" qualification comparator that's necessary to exclude people who have a qualification.

Sarah

Dan

Amanda

Zach

Linda

Maryam