Kathleen Currie Hall

Notes on: Pater, Joe, Christine Stager, and Janet F. Werker. 2004. The perceptual acquisition of phonological contrasts. Language, 80(3): 384-402.

Summary:
In this article, the authors extend a series of perceptual experiments with 14-month-olds, testing the difference between perception of phonetic distinction vs. phonological contrast.  They find that even though children can hear distinctions between minimally different words in terms of voicing and place, when these words are actually paired with objects and the child tries to attach meaning to them, the phonetic differences are not noticed.

Details:
1. Two hypotheses about the acquisition of perception: (a) because children start out being able to discriminate all possible contrasts, they must have very detailed phonetic representations of the words they hear and store in their mental lexicon.  (b) as with production, perception proceeds gradually, and some contrasts are easier to perceive than others; even older kids can't always perform well on minimal word decision tasks.
2. They use a new-ish technique, the SWITCH method, to test these hypotheses.  In this technique, the child is presented with a picture of an object accompanied by a spoken word, and is allowed to habituate to the pairing.  In the test trial, the child is presented with either the same pairing or a "switched" pairing (e.g., same picture, different word), and the duration of the child's interest is measured.  It is assumed that if the child pays attention longer to the switched pairing, it is because the child noticed the difference and is interested in this new information.  This technique can also be used with two object-word pairings, where the "switch" reverses the pairings (a more difficult task).
3. Exp. 1: Replicate earlier double-object switch experiment using [bIn] and [dIn] instead of [bI] and [dI] to make sure that the fact that infants didn't notice the switch wasn't just because the words weren't English.  Replicated the earlier results: the children didn't notice the switch -- seem to perceive [bIn] and [dIn] as the same when learning them in a sound-object pairing.  (Note that earlier experiments have shown that when just listening to the sounds, infants this age CAN discriminate these sounds -- it seems to be an effect of the sound-meaning pairing that impairs discrimination.)
4. Exp. 2: Single-object pairing, voice contrast instead of place contrast ([bIn] vs. [pIn]).  Same results: infants don't notice the difference.
5. Exp. 3: Single-object pairing; both voice and place contrast ([dIn] vs. [pIn]).  Again, infants don't notice the difference.
6. Two main possible interpretations: (a) children actually encode these words e.g. [bin] vs. [pin] as same in their mental lexicons. (b) children are too busy trying to pair the sound with the meaning to notice the distinction -- an effect of processing only.  The experiments presented here did not differentiate between these two interpretations.  Either way, somehow the phonetic differences are not fully utilized in the phonological system.
7. BUT -- other similar experiments with words that the child knows indicate that the child _can_ pay attention to the difference!  So the differences are only _partially_ encoded.
8. The authors (who favour the "processing overload" interpretation) claim that the phonological grammar of the child can be modelled in OT by having markedness constraints outrank faithfulness constraints (faithful = faithful to phonetic input of signal) by default.  At first, and whenever processing difficulty is high, this pattern reigns, and phonetic differences are neutralized in the phonological representations.  But when processing is easier for whatever reason, then faithfulness can take over and the contrast can be maintained.

Questions:
1. p. 390 (and others) -- what qualified as "exposure" to one of the test words?  Surely it's impossible for the parents to recall everything said in the child's presence in 14 months . . . !
2. Not sure I buy their OT analysis . . . I like the basic idea (modelling the actual behaviour of the child with your phonological theory, the fact that it gives you the gradual/partial integration, etc.).  But it seems intuitively very wrong.  I understand the basic principle of needing to have markedness over faithfulness in OT learning systems, but here it seems very strange when applied to the phonetic input to the child.  Essentially, isn't the baby saying (not consciously of course!) "Oops, every time I'm under a lot of stress, I keep forgetting that I should be maintaining whatever sounds I hear!"  Why should the child posit anything other than the surface form that they hear?  (Seems strange that the default is to in many ways be _more_ abstract . . . .)