Allophonic variation and the acquisition of phoneme categories
Q: How do children acquire the distinction between allophones and phonemes?
H: A minimal pair analysis (would need a substantial lexicon) or a distributional analysis.
So, actually, the authors don't directly address this question, but this is the big question which motivates the other questions that they do directly attempt to answer.
Experiment 1: Perception of Phonemic vs Allophonic Contrasts
Q: How well is the allophonic contrast perceived? Does it vary in isolation versus in a phonological context?
Q: Is the perception of the contrast affected in a phonotactically legal vs illegal string?
Method:12 Native French speakers (no ages given) listen to [VC.CV] in which the first syllable is a/i/u + voiced uvular fricative or voiceless uvular fricative OR m/n and the second syllable is an obstruent + a/e/i/o/u
- The voiced/voiceless uvular fricative is allophonic in French.
- The m/n option is to compare phonemic/allophonic contrasts.
- The adjacent consonants agree in voicing
French speakers are told [VC.CV] is a two word phrase in a foreign language and are asked if the first word is the same. They also hear the first syllable in isolation and are asked if it is the same.
Results: ANOVA- discrimination is better without context p<.02, in context, discrimination of m/n is better than the uvular fricatives p<.02
So, the uvular fricatives are acoustically different and an be perceived in isolation, but in the phonological context the discrimination of the allophonic contrast differs significantly from that of a phonemic contrast.
Experiment 2: Acquisition of Phoneme Categories
Q: Does the same effect arise with the allophonic contrast in context?
Q: Is the bias to create two categories given a bimodal distribution overriden if they appear in complementary distribution?
P: Monomodal distribution--> 1 category, Bimodal Distribution--> 2 categories, Bimodal distribution+ complementary distribution--> 1 category
Method: 60 native French speakers are given exposure in each of the three patterns listed in the predictions. Tokens are a/i/u+ a continuum from the voiced to the unvoiced uvular fricative.
Results: ANOVA- difference between pre and post tests was significant only in the bimodal group; no block (pre vs post test) x group (distribution) interaction
They draw pretty strong conclusions (that the results are in accordance with the predictions) from this statistically very weak data. I found the results to be dubious.
My Reaction:
I found this paper to be more difficult to read (and less interesting) but I'm glad I waited until after I discussed it with my advisor to write this. Most of my qualms came from the second experiment, but my advisor's primary interest is with the first.
The question is why there is increased difficulty of perception of the allophones in context. The paper offers three options:
- The first they dismiss off hand is a low-level masking explanation, which they say is unlikely because that explanation would seem to affect the phonemic contrast as well.
- The second is that allophonic contrasts are initially represented as two distinct category and then if they are in the right context, there is some 'inverse phonology' which recodes the segments as being the same. This they also reject, because this predicts that in illegal contexts, the allophones should be easily discriminated, which didn't happen in their experiment.
- The third is that allophones are represented as a single phonological category in a similar way to that of nonnative contrasts being mapped to a single category.
My advisor noted that low level masking explanations are often pooh-poohed but in her opinion, there may be some merit to an explanation like this, particularly if there is something "special" about allophone perception.