[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: perception of durational variability



Hi Volker:

O.K., if this is really your hypothesis (I am referring to your last paragraph), I wish you good luck. Perhaps you have discovered an important confound here. However, people also have stereotypes (which may or may not be accurate) about the speaking rates of different languages. So, speeding German up may well make it sound more like French in terms of rate, but this leaves open the question of whether it sounds more like French in terms of temporal regularity.

Your problem reminds me of the difficulty musically untrained listeners have in judging the relative pitch of tones differing in timbre. They are instructed to judge pitch, but they cannot avoid being influenced by timbre. Your listeners seem to have difficulty judging temporal regularity independent of rate. Perhaps you need some expert listeners, such as musicians or phoneticians, who really can keep these dimensions apart. Also, perhaps your response categories should not be simply "French" and "German" but "Fast French", "Slow French", "Fast German", and "Slow German".

Best,
Bruno

Hi Bruno,

Thanks for the detailed explanation. Did your participants rate the slower rate stimuli as more regular, or the faster ones?

The faster ones were rated more regular.

In any case, the problem with your stimuli may be that they were all fairly irregular,

Correct, but this, I find, is exactly what happens in speech. My primary interest is not to investigate how subjects perceive irregularity of intervals in general but how they perceive the irregularity differences of c and v intervals typically occurring in different languages.


so participants may have relied on rate because it was a much more salient property.

Correct as well, and, again, I assume that this may be the case in speech too.

You may have to prevent this by giving more precise instructions and/or controlling the rate of your stimuli.

Well, it has been shown already that when subjects are trained on the stimuli and when rate is controlled for they can use variability cues without a problem. I deliberately left the interpretation of 'regularity' to my subjects and claim that when they have the choice they choose rate rather than interval variability (or rather: the degree of interval variability occurring in speech) to decide whether a sequence of variable intervals is more or less regular. Again, if listeners should use these cues in real situations to distinguish between languages nobody will tell them which cues to listen for and speech rate will not be normalized.


It seems unlikely to me that linguists' classification of languages as stress- or syllable-timed merely reflects a difference in the average speaking rate for those languages.

Well, generally this is my idea. I would not go as far as saying that it 'merely' reflects speech rate, but I do claim that speech rate, amongst other factors, plays a major role in there which has so far been entirely neglected. I am currently running experiments with delexicalised stimuli that are closer to real speech (low pass filtered speech at 300 Hz) and I am manipulating the rates of stimuli in order to test whether I can make listeners rate German stimuli as French based on rate only.


Best wishes,
Volker







Best,
Bruno

Hi Bruno,

I don't think there are any data suggesting that people cannot distinguish interval durations at fast rates. The question is how large the differences must be to be detected, and how the magnitude of that difference depends on rate. There is no "breakdown" of discrimination at any rate.

Yes, thanks for pointing that out so clearly. After reading through the literature this morning I realized that I got something wrong there...

It is unclear what your listeners had to do. You are talking about "a
big effect of rate on listeners' perception of speech rhythm", but what does this have to do with interval discrimination? What exactly
was the effect of rate on perceived speech rhythm, and how did interval durations vary?

Well, sorry, it was indeed vague. I was not sure to what level of
detail readers on this list would be interested in this. I was referring to recent theories of speech rhythm claiming that rhythm classes (e.g. stress- and syllable-timed languages) can be distinguished by the listener based on the durational variability of consonantal (c) and vocalic (v) intervals (e.g. work by Frank Ramus or Esther Grabe). It was
demonstrated that syllable-timed languages for example have proportionally less c- and v-interval variability than stress-timed languages and that this information is processed by the listener to group languages into traditional rhythmic classes. For two syllable-timed (French & Italian; F & I) and two stress-timed languages (English & German; E & G) my own data replicates the objective differences nicely. However, I found that because of their less complex syllable structure, speakers of syllable-timed languages also produce cv-intervals at a far higher rate than stress-timed languages.


In a perception experiment I took sentences from French and German and turned v-intervals into tones and c-intervals into white noise and asked listeners to rate the stimuli according to the 'regularity of beep sequences' on a scale form 1 to 10. I left the interpretation of 'regularity' to the listener and expected that listeners would pick up on the proportionally higher c- and v-interval variability in German and thus rate these stimuli as the more irregular beep sequences. However, results showed very poor correlation between the regularity rating and any of a number of c- and v-variability measures but I found a strong correlation with cv-rate. So it seems that in my experiment listeners interpreted 'regularity' as rate only and did not listen for any durational variability within the stimuli.

In order to interpret the results I thought it may help to consult work
that looked at the influence of rate on interval variability perception. However, I find my data and method are pretty difficult to compare with Friberg & Sundberg and the type of studies mentioned in there. If I had pointed out to my listeners to listen out for certain types of durational variability I am sure they could have done it (Ramus showed that in a way by not allowing rate variability). However, the fact that they do not make use of durational variability cues when given the choice between rate and variability tells me that this may be something they do in speech too when they distinguish between languages based on rhythmic cues.


Hope that makes it clearer,
Volker






Best, Bruno

Bruno and Pierre,

thank you so much for your helpful suggestions!

The work on rhythm is more what I am looking for. I found a big effect of rate on listeners' perception of speech rhythm. I assume
that it may have something to do with listeners not being able to
detect interval variability in speech any more when the intervals
under investigation are shorter (typically the case in so called 'syllable-timed languages' because they posses simpler phonotactic
structures). So I am looking for evidence showing at what rate interval distinction ability breaks down in rhythmic contexts.


However, all interval durations I am looking at (syllables, c- or v-intervals) are well below 200 ms in any language I have collected
data on, which, judged by the rhythm findings, would mean that listeners should not be able to detect durational variability at all between any of the speech intervals (when judging duration only!) and that can hardly be true. It probably has to do with the
fact that interval variability in my speech stimuli is much more complex and do not fulfill the criterion of isochrony in the way they do it in the Friberg & Sundberg study. I am working on an explanation...


Best wishes & thanks again, Volker

-- -------------------------------------------- Volker Dellwo Department of Phonetics & Linguistics University College London

phone: +44 (0)20 7679 5003 (internal: 25003)

www.phon.ucl.ac.uk www.phonetiklabor.de --------------------------------------------




--
--------------------------------------------
Volker Dellwo
Department of Phonetics & Linguistics
University College London

phone: +44 (0)20 7679 5003 (internal: 25003)

www.phon.ucl.ac.uk
www.phonetiklabor.de
--------------------------------------------




--
--------------------------------------------
Volker Dellwo
Department of Phonetics & Linguistics
University College London

phone: +44 (0)20 7679 5003 (internal: 25003)

www.phon.ucl.ac.uk
www.phonetiklabor.de
--------------------------------------------


--
Bruno H. Repp
Haskins Laboratories
300 George Street
New Haven, CT 06511-6624
Tel. (203) 865-6163, ext. 236
Fax (203) 865-8963
http://www.haskins.yale.edu/staff/repp.html