Perception on complex onsets

[Joe Stemberger]

Hi, Yvan.

I presented a paper on cluster perception in adult English about 20 
years ago, but it has never been written up and published (the first 
author went off into industry and took the poster materials with him, 
so...):

Yuchtman, M., Stemberger, J.P., & Martin, C. (1987). "Recognition of 
consonant clusters". Paper presented at the meeting of the Acoustical 
Society of America, Miami, November.

We had all possible English onsets followed by the vowel /a/ with no 
coda:  /ba, bla, bra, pa, .../
People were free to wrote down exactly what consonants they heard (if any).
We looked at the speech in quiet (as recorded, with stimuli normalized 
for loudness), and with signal-to-noise ratios of +10dB, +5dB, and 0dB.

I went back and looked at the results with your /gl/ question in mind.

There basically is no effect of place or voicing in stops in quiet, 
+10dB, or +5dB --- except for a lot of error on /bla/ (perceived as /la/ 
or less commonly /wa/). No special error rates on velars.

At 0dB, though, the error rates shot way up for all clusters starting 
with /g/ (/gla, gra, gwa/), and for /dwa/; errors in which subjects 
reported just the second consonant accounted for about 25% of trials.
Similar errors for /kla, kra, kwa/ occurred at about 5% of trials.
(And /twa/ was perceived as /wa/ 12% of the time.)
And rates of similar errors on /bra/ and /dra/ were also at 5% (though 
/bla/ also had 15% of trials perceived as /wa/).

So there wasn't anything to pull out /gl/ as special, but a general 
problem with /g/, plus a few other difficult clusters.

Of course, the phonetics of English /kl/ and /gl/ is different from in 
French, so I'm not sure how well this generalizes.


On the non-perceptual side:
Voicing is phonetically hardest to produce in velars, and it's common 
for English-learning children to have [k] but not [g].
Even if this child has [g] for singleton onsets, perhaps the reduction 
of this cluster to [l] might be a reflection of the difficulty of [g]?


---Joe Stemberger
Linguistics
UBC



Yvan Rose pravi:
> Dear Info-CHILDES members,
> We are currently looking at data that suggest an asymmetry between the 
> behaviours of /kl/ and /gl/ clusters in a young French-learning child. 
> The asymmetry goes as follows:
> 
> /kl/ -> [k]/[kl]
> /gl/ -> [l]
> 
> These data are difficult to explain from a representational perspectives 
> (the clusters should share similar prosodic representations) and cannot 
> be explained through a statistical account. We are now entertaining the 
> possibility that perceptual factors may play a role.
> 
> If you could suggest references about the perception of such clusters, 
> or of velars and liquids in general, we would most appreciate it. We 
> will of course collate the results and report them back to the list for 
> your interest.
> 
> Thank you very much in advance,
> Yvan Rose (& Christophe dos Santos)
> 
>