Correlates of linguistic rhythm in the speech signal

TL;DR: In this article, the authors present instrumental measurements based on a consonant/vowel segmentation for eight languages and show that intuitive rhythm types reflect specific phonological properties, which in turn are signaled by the acoustic/phonetic properties of speech.

About: This article is published in Cognition.The article was published on 1999-12-17 and is currently open access. It has received 1168 citations till now. The article focuses on the topics: Phonetics & Psycholinguistics.

Summary

1 Introduction

• There is a clear difference between the prosody of languages such as Spanish or Italian on the one hand and that of languages like English or Dutch on the other hand.
• Thus, precocious detection of the rhythm type of their native language might be a simple way for infants to decide which representation unit to use for further speech analysis (this view is further discussed in the general discussion).
• 2 Current views on speech rhythm 2.1 Against the isochrony theory Given the excellent reasons for believing in rhythmic classes, one would expect that these groups of languages should differ by readily identifiable acoustic or phonetic parameters.
• This is not the picture provided by past studies on rhythm.
• The results of this research are a) that variation in syllable duration is similar in all six languages and b) that stress pulses are not more evenly spaced in the second group of languages than they are in the first.

3.2 Material

• Sentences were selected from a multi-language corpus initially recorded by Nazzi et al. (1998) and augmented for the present study (Polish and Catalan4).
• ∆V suggests that there may be more to speech rhythm than just these distinctions; this variable, although correlated with the two others, rather emphasizes differences between Polish and the other languages.
• At this stage, new discrimination experiments are clearly needed to test whether ∆V plays a role in rhythm perception.
• 4 Confrontation with behavioral data Following Bertinetto (1981), Dasher and Bolinger (1982) and Dauer (1983), the authors have assumed that the standard rhythm classes postulated by linguists may be the result of the presence and interaction of certain phonological features in languages.
• Others have managed to isolate intonation by producing a tone following the fundamental frequency of utterances (de Pijper, 1983; Maidment, 1976; 1983; Willems, 1982).

4.1.2 Modeling the task

• The procedure used by Ramus & Mehler consisted in training subjects for L1/L2 categorization on 20 sentences uttered by 2 speakers per language, and then having them generalize the categorization on 20 new sentences uttered by 2 new speakers.
• Given a numerical predictor variable V and a binary categorical variable L over a number of points, this statistical procedure finds the cut-off value of V that best accounts for the two values of L. Straightforwardly, the authors take language, restricted to the English/Japanese pair, as the categorical variable, and %V as the numerical variable.
• %V was chosen rather than ∆C because it has lesser variance and therefore can be expected to yield cleaner results.
• The simulation will thus include the same number of sentences as the behavioral experiment.

4.2.2 The discrimination task

• Discrimination studies in infants report two kinds of behavior: first, recognition and/or preference for maternal language, and second, discrimination of unfamiliar languages.
• It requires, however, familiarization with one language during the experiment, as is done in habituation/dishabituation procedures.
• Thus, in both cases, discrimination behavior involves forming a representation of one language, and comparing utterances from the new language with this representation.
• For this reason, neither standard comparisons between sets of data nor procedures involving supervised training (like the logistic regression) can adequately model the task of the infant, since they would a priori presuppose two categories, when infants discover by themselves that there are (or not) two categories.
• For this purpose the authors will model infants' representation of sentences' rhythm, their representation of a language, and their arousal in response to sentences.

4.2.3 A model of the task

• Below are the main features and assumptions of the model: Here, the authors note steps with the index n. the rhythm of each sentence.
• The authors further assume that there is a causal and positive correlation between arousal and sucking rates observed in experiments, that is, a rise in arousal causes a rise in sucking rate.
• There are important differences between the proposed simulations and the real experiments that deserve to be discussed: in the experiments, switch to the test phase follows reaching a certain habituation criterion, namely, a significant decrease in the sucking rates.
• In most experiments, the stimuli used consisted in longer samples of speech in each language than the authors have here.

4.2.4 Results

• Simulations were run on all 26 pairs of languages studied in this paper.
• Four language pairs (marked by asterisks in the table) present a peculiar arousal pattern, in that the control group has a higher average arousal in the test phase than the experimental group.
• Only 2 pairs of languages do not conform to this pattern: Polish/French and Spanish/Dutch, for which no discrimination is predicted by the simulation.
• It seems to us that they are not completely compatible with the otherwise high coherence of their data.

4.2.5 Groups of languages

• Nazzi, Bertoncini and Mehler (1998) have also tested discrimination between groups of languages (Table 3), and the authors have tried to simulate this experiment as well.
• Indeed, subjects in the control group are presented a different combination of languages than subjects in the experimental group.
• From Figure 1 the authors can guess that ∆C would have predicted the same pattern of results, but simulations might have been less sensitive.
• 5 General discussion Phonetic science has attempted to capture the intuitive notion that spoken languages have characteristic underlying rhythmic patterns.
• Given all these considerations, the authors believe that the notion of three distinct and exclusive rhythm classes has not yet been definitively proven, but rather is the best description of the current evidence.

100 for ease of reading.

• Scores are classification percentages on the test sentences obtained from logistic regressions on the training sentences.
• In these cases one of the two regressions failed to converge, meaning that the solution of the regression was not unique.
• Figure Captions Figure 1. Distribution of languages over the (%V, ∆C) plane.
• Figure 6. Simulated arousal pattern for English/Spanish discrimination.

Figures

Figure 1

Figure 6

Figure 5

Figure 4

Figure 7

Figure 3

Figure 2

Frequently asked questions

Q1. What contributions have the authors mentioned in the paper "Correlates of linguistic rhythm 1 running head: correlates of linguistic rhythm correlates of linguistic rhythm in the speech signal" ?

This paper presents instrumental measurements based on a consonant/vowel segmentation for eight languages. The measurements suggest that intuitive rhythm types reflect specific phonological properties, which in turn are signaled by the acoustic/phonetic properties of speech. 

Q2. How many sentences of each language will be used as training set?

10 sentences of each language, uttered by 2 speakers per language, will be used as training set, and the 10 remaining sentences per language, uttered by other speakers, will be used as test set. 

Q3. Why do infants seem to discriminate only between native and foreign languages?

after 2 months, infants seem to discriminate only between native and foreign language, due to early focusing on their native language (Christophe & Morton, 1998; Mehler et al., 1988). 

Q4. How did the authors get the 95% hit rate in the test phase?

To assess any asymmetry between the training and the test sets, the authors redid the regression after exchanging the two sets, and the authors obtained a 95% hit rate in the test phase (chance is 50%). 

Q5. What are some properties that have been proposed to be more or less connected with rhythm?

A number of properties have been proposed to be more or less connected with rhythm: vowel reduction, quantity contrasts, gemination, the presence of tones, vowel harmony, the role of word accent and of course syllable structure (Dauer, 1987; Donegan & Stampe, 1983; see Auer, 1993 for a survey). 

Q6. What did Bolinger show about the duration of interstress intervals?

Bolinger (1965) also showed that the duration of interstress intervals is influenced by the specific types of syllables they contain as well as by the position of the interval within the utterance. 

Q7. What are the factors that influence rhythm?

Only vowel reduction and contrastive vowel length have been described as factors influencing rhythm (Dauer, 1987), but the present analysis suggests that the other factors may do so as well. 

Q8. What is the likelihood that the properties underlying the classes would be able to be set?

if all languages could be sorted into a few rhythm classes, the likelihood that the properties underlying the classes might be cues that allow for the setting of grammatical parameters would be increased. 

Q9. What is the reason why there is no space for a distinct class?

Since languages of the Unmarked and Marked II types are not part of their corpus, the authors cannot assess the relevance of these two additional classes, but it is a fact that on Figure 1 for instance, there seems to be space for a distinct class between Catalan and Japanese, and of course there's also space for another class beyond Japanese.