Elsevier

Journal of Phonetics

Volume 38, Issue 3, July 2010, Pages 422-430
Journal of Phonetics

Perception of predictable stress: A cross-linguistic investigation

https://doi.org/10.1016/j.wocn.2010.04.001Get rights and content

Abstract

Previous studies have documented that speakers of French, a language with predictable stress, have difficulty distinguishing nonsense words that vary in stress position solely (stress “deafness”). In a sequence recall task with adult speakers of five languages with predictable stress (Standard French, Southeastern French, Finnish, Hungarian and Polish) and one language with non-predictable stress (Spanish), it was found that speakers of all languages with predictable stress except Polish exhibited a strong stress “deafness”, while Spanish speakers exhibited no such “deafness”. Polish speakers yielded an intermediate pattern of results: they exhibited a weak stress “deafness”. These findings are discussed in light of current theoretical models of speech perception.

Introduction

It is well-known that adult listeners experience difficulty in perceiving phonological contrasts that are not used in their native language. For instance, Japanese has only one liquid consonant, /ɾ/, and Japanese listeners have difficulty perceiving the contrast between the English liquids /r/ and /l/ (Goto (1971), Miyawaki et al. (1975)), even if they are learners of English (Guion, Flege, Akahane-Yamada, & Pruitt (2000), Logan, Lively, & Pisoni (1991), Takagi & Mann (1995)). Similar difficulties have been reported for other cases of non-native consonantal and vocalic contrasts (Pallier, Bosch, & Sebastián-Gallés (1997), Werker & Tees (1984a)). Moreover, they are known to emerge during the first year of life (Werker & Tees (1984b), Werker & Tees (1999)). From a theoretical viewpoint, difficulties with the perception of non-native contrasts shed light on an important question: how does linguistic knowledge impact on our speech perception system?

Two types of accounts have been proposed regarding language-specific effects in the perceptual processing of consonants and vowels. We call them functional role accounts and lexical statistics accounts, respectively. In functional role accounts, dimensions that play a functional role in the language are amplified, and dimensions that are non-functional are attenuated, resulting in good perception of the former and poor perception of the latter. In several speech perception models, dimensions are expressed either in terms of acoustic or phonetic cues (Francis & Nusbaum (2002), Iverson et al. (2003), Jusczyk (1997)) or in terms of phonological features (Brown (1998), Brown (2000), Flege (1995), Lado (1957)). Lexical statistics accounts, by contrast, posit that the phonological grammar emerges from generalizations about phonological regularities across the lexicon. In particular, the more regular or predictable a pattern is, the less it needs to be specified in the lexical representation. For instance, if vowel nasality is entirely predictable, as in English where nasal vowels only occur before nasal consonants, it does not need to be specified lexically; similarly for predictable stress. Two types of lexical statistics accounts are exemplar-based models (Pierrehumbert (2001), Pierrehumbert (2003)) and connectionist models (Gaskell (2003), Gaskell, Hare, & Marslen-Wilson (1995)).

Suprasegmental contrasts (i.e., contrasts concerning tone, length, stress, and pitch accent) likewise give rise to perceptual difficulties in speakers of languages that lack such contrasts. For instance, English speakers have difficulty perceiving the tonal contrasts of Mandarin (Gottfried & Suiter (1997), Hallé, Chang, & Best (2004), Kiriloff (1969)), and French speakers have difficulty perceiving the vowel length contrast of Japanese (Dupoux, Kakehi, Hirose, Pallier, & Mehler, 1999). French speakers also have difficulty perceiving the stress contrast of Spanish (Dupoux, Pallier, Sebastián, & Mehler (1997), Dupoux, Peperkamp, & Sebastián-Gallés (2001)), even if they are advanced learners of Spanish (Dupoux, Sebastián-Gallés, Navarrete, & Peperkamp, 2008). These difficulties with perceiving suprasegmental contrasts, like those concerning segmental contrasts, arise early: At nine months of age, English-learning infants already differ from Chinese ones and fail to discriminate a tonal contrast (Mattock & Burnham, 2006). Likewise, French-learning infants of the same age differ from Spanish-learning infants and fail to discriminate a stress contrast (Skoruppa et al., 2009).

Unlike in the case of segmental contrasts, little is known about the mechanisms that yield difficulties perceiving suprasegmental contrasts. In this article, we address this issue in a cross-linguistic experiment concerning the perception of stress. Following our previous work on this issue (Dupoux, Pallier, Sebastián, & Mehler (1997), Dupoux, Peperkamp, & Sebastián-Gallés (2001), Dupoux, Sebastián-Gallés, Navarrete, & Peperkamp (2008), Peperkamp & Dupoux (2002)), we use the shorthand stress “deafness” to refer to the difficulty with the perception of stress at a phonological level.

Besides speakers of French, we test speakers of four more languages with predictable stress, i.e. Southeastern French (which – as we will show below – differs from Standard French in the distribution of stressed and unstressed syllables), Finnish, Hungarian, and Polish. As in our previous work, our control population consists of speakers of Spanish, whose language has contrastive stress, as evidenced by minimal pairs such as /ˈbebe/ “(s/he) drinks” – /beˈbe/ “baby”. Our aim is to identify which factor or factors govern(s) the emergence of stress “deafness”. We begin by reviewing the predictions of functional role accounts and lexical statistics accounts for the perception of stress.

As indicated above, in functional role accounts the weighting of acoustic/phonetic cues or phonological features can be reduced or enhanced, depending on their functional roles in the language (Brown (1998), Brown (2000), Flege (1995), Francis & Nusbaum (2002), Iverson et al. (2003), Jusczyk (1997), Lado (1957)). It should be noted, however, that contrary to segments, suprasegmentals may have functional roles at more than just the lexical level. For instance, stress is instantiated by three phonetic cues: duration, F0 and intensity (Lehiste, 1970). In addition to lexical stress, these cues can signal prosodic constituent boundaries and grammatical functions (interrogative versus affirmative mode, focus, etc.). These prosodic and grammatical levels should clearly not be included in the definition of functional role, since in French, duration, F0 and intensity are known to signal boundaries and to serve grammatical functions; hence, it is only by restricting the definition of functional role to the lexical level that we can account for the reduction of sensitivity to stress cues in French speakers. Under this definition of functional role, two factors can be identified that might play a role in the emergence of stress “deafness”:

The first factor concerns the domain of stress assignment. Stress is typically a word-level property, in that all content words have one main stress. French (both its standard and its Southeastern variety) is an exception, in that stress is argued to be assigned at the phrase level (Grammont, 1965). Perhaps this is what induces stress “deafness” in French as opposed to Spanish speakers: they can completely ignore stress for the purposes of word recognition. Consequently, we would expect speakers of Finnish, Hungarian, and Polish, which all have a word stress rule, to be like speakers of Spanish and exhibit no stress “deafness”, while speakers of Southeastern French should be like speakers of Standard French and exhibit stress “deafness”.

The second factor concerns the lexical use of one or more phonetic correlates of stress. Duration and F0 are not only phonetic correlates of stress but are also used for other suprasegmentals: duration is the phonetic correlate of contrastive length, while F0 is the correlate of tone and pitch accent. Among our test languages with predictable stress, Finnish and Hungarian have contrastive vowel length (Karlsson (1999), Tompa (1972)). Thus, Finnish and Hungarian speakers use duration lexically. French and Polish speakers, by contrast, do not use any of the stress cues lexically, since their languages do not make lexical use of length, tone, or pitch accent. Maybe the perception of stress correlates with the lexical use of stress cues in the listener's native language. We should then obtain three levels of stress perception: Standard French, Southeastern French, and Polish speakers should exhibit a strong “deafness” (no lexical use of stress cues at all), Finnish and Hungarian speakers a weak “deafness” (lexical use of duration for contrastive vowel length), and Spanish speakers no stress “deafness” at all (lexical use of duration, F0 and intensity for contrastive stress). An additional prediction is that when duration is not part of the cues by which stress is realized, the performance of Finnish, Hungarian, and Spanish speakers should decrease more than that of Standard French, Southeastern French and Polish speakers. Indeed, whereas all populations should have more difficulties when one of the phonetic correlates is absent, this disadvantage should be larger for those who make lexical use of the missing correlate in their native language.

Concerning lexical statistics accounts, we mentioned above that they focus on the existence of phonological distributional regularities across lexical items, with more regular patterns being encoded less precisely (Gaskell (2003), Gaskell, Hare, & Marslen-Wilson (1995), Pierrehumbert (2001), Pierrehumbert (2003)). For these accounts, two different factors can be identified that might play a role in the emergence of stress “deafness”.

First, languages differ with respect to the variability in the position of stress. In Standard French, stress surfaces at a fixed position within its phrasal domain: stress always falls on the phrase-final syllable (even when the phrase ends in a function word). This is where Standard French diverges from Southeastern French. In Southeastern French, phrases can end in the vowel schwa, which surfaces unstressed (Durand, 1976). Consequently, stress is not on a fixed position in Southeastern French: it is final unless the final vowel is schwa, in which case it is penultimate. As to the other languages with predictable stress, stress is assigned within words. Finnish and Hungarian have stress on a fixed position within this domain, namely the initial syllable (Karlsson (1999), Tompa (1972)). In Polish, by contrast, stress is not on a fixed position. Specifically, stress is regularly assigned to the word's penultimate syllable (Comrie, 1967), but not all words end in a sequence of a stressed syllable followed by an unstressed one. This is because Polish has a large number of monosyllabic content words, which are stressed too. Finally, although Spanish has contrastive stress, the position of stress is confined to one of the word's last three syllables (Harris, 1983). Maybe it is the complete absence of variability in the position of stress that yields stress “deafness” in speakers of Standard French. In that case, we expect the same result for speakers of Finnish and Hungarian, but less difficulties with the perception of stress in speakers of Southeastern French and Polish. The position of stress being the most variable in Spanish, speakers of this language should, moreover, outperform speakers of all other languages.

Second, languages differ with respect to the presence of lexical exceptions. Indeed, some languages with predictable stress have lexical exceptions to their stress rule, others do not. French is an example of the latter: all utterances end in a stressed syllable, even if the last word is a loanword. Southeastern French, Finnish and Hungarian are also exceptionless. Polish, however, does have a small number of exceptional words with irregular stress on either the antepenultimate or the final syllable (Comrie, 1967). These words are most often words of Greek or Latin origin (gimˈnastyka) or foreign proper names (ˈEverest).1 Finally, even though Spanish is described as having contrastive stress, it has a default stress pattern: in polysyllables, stress tends to fall on the penultimate syllable if the word ends in /n/, /s/ or a vowel, and on the last syllable otherwise (Harris, 1983). In the Spanish database of Santiago, Justicia, Palma, Huertas, and Gutiérrez (1996), 83% of the lexemes obey this default pattern (we consider monosyllabic content words to have regular stress, regardless of their final segment). Hence, we might re-describe Spanish as having predictable stress with 17% of lexical exceptions. Obviously, the more numerous the lexical exceptions are, the more advantageous it is to reliably encode stress in the phonological representation. This yields the prediction that there are three levels of performance as far as stress perception is concerned: a strong “deafness” should be found in speakers of Standard French, Southeastern French, Finnish and Hungarian, a weak “deafness” in Polish speakers, and no stress “deafness” in Spanish speakers.

To summarize, our six test languages vary with respect to four factors that might influence stress perception (see Table 1): (1) the domain of stress; (2) the lexical use of stress cues (for contrastive stress or some other suprasegmental); (3) the variability in stress position; and (4) the existence of lexical exceptions.

A summary of the predictions is shown in Table 2. It should be noted that no two factors make the same predictions.

Stress “deafness” in French speakers has been observed both in an ABX discrimination task (Dupoux et al., 1997) and in a sequence recall task (Dupoux et al., 2001). In order to assess the perception of stress by speakers of the six languages in Table 1, we use the sequence recall task, which has been shown to provide a more robust indicator of stress “deafness”. Indeed, it was found to yield a complete separation between French and Spanish monolinguals (Dupoux et al., 2001). Moreover, it is largely insensitive to experience effects, such as the late acquisition of Spanish by French speakers (Dupoux et al., 2008). In this paradigm, participants have to recall sequences of two auditorily presented non-words that differ either in the position of stress or in the middle consonant, with the various tokens of each non-word being phonetically different. The combination of a memory load and phonetic variability ensures that this task taps an abstract phonological processing level, and cannot be performed using a low-level acoustic strategy.

In Dupoux et al. (2001), speakers of Standard French but not those of Spanish were found to exhibit stress “deafness”: they had severe difficulty perceiving the stress contrast compared to the consonantal contrast. The same paradigm was used by Peperkamp and Dupoux (2002) in two experiments testing, respectively, (Standard) French, Finnish and Spanish (Experiment 1), and Hungarian and Polish (Experiment 2). However, this study did not allow for firm conclusions regarding the effect of the four factors outlined above, for the following reasons. As shown in Table 2, Factors 1 and 2, unlike Factors 3 and 4, predict a stronger stress “deafness” in Standard French speakers than in Finnish and Hungarian speakers. Peperkamp and Dupoux reported an inconclusive, marginal, difference in the expected direction between Standard French and Finnish speakers, and a significant difference – also in the expected direction – between French and Hungarian speakers. However, the latter difference was difficult to interpret, due to the fact that different sets of stimuli and slightly different procedures were used to test the French and Hungarian groups; most importantly, the strength of the durational cue and the length of the test sequences differed. Table 2 also shows that Factors 2, 3, and 4, unlike Factor 1, predict a difference between Polish and Spanish speakers. Peperkamp and Dupoux found that Spanish speakers perceive stress better than Polish speakers, but again, this effect was only marginally significant and the stimuli and procedure used to test the Polish and the Spanish groups were different.

In the present study, we report on a novel perception experiment with the sequence recall task, which uses a single set of stimuli across all languages. To test the effect of the presence versus absence of the durational cue (see Table 2, Factor 2), there are two versions of the stress contrast: in one version, stress is instantiated by all three stress cues (duration, F0, and intensity); in the other one, the durational cue is removed.

Section snippets

Experiment

We use a variant of the sequence recall task described in Dupoux et al. (2001). The experiment is divided into three parts. In each part, participants are required to learn two CVCV non-words that are a minimal pair differing only in one phonological characteristic: place of articulation of the middle consonant in the first part, and location of stress in the second and third part. In each part, participants are taught to associate the two non-words to the keys [1] and [2], respectively, of a

General discussion

Using a short-term memory sequence recall task, we found that speakers of four languages with predictable stress, namely, Standard French, Southeastern French, Finnish and Hungarian, exhibit a strong stress “deafness”; that is, they have difficulty perceiving a stress contrast as compared to a phoneme contrast. The “deafness” effect is numerically large and does not differ among the languages. These results clarify the potential difference between French and Hungarian speakers found previously (

Acknowledgments

This research was supported by grants from the Centre National de la Recherche Scientifique (Aide à Projet Nouveau, and Cognition et Traitement de l’Information, CTI02-15), the Agence Nationale pour la Recherche (ANR-05-BLAN-0065-01), and the European Union (FP6 012738). We would like to thank Anne Christophe for comments and discussion.

References (42)

  • C. Brown

    The role of the L1 grammar in the L2 acquisition of segmental structure

    Second Language Research

    (1998)
  • C. Brown

    The interrelation between speech perception and phonological acquisition from infant to adult

  • B. Comrie

    Irregular stress in Polish and Macedonian

    International Review of Slavic Linguistics

    (1967)
  • E. Dupoux et al.

    Epenthetic vowels in Japanese: A perceptual illusion?

    Journal of Experimental Psychology: Human Perception and Performance

    (1999)
  • E. Dupoux et al.

    A robust paradigm to study stress “deafness”

    Journal of the Acoustical Society of America

    (2001)
  • J. Durand

    Generative phonology, dependency phonology and southern French

    Lingua e Stile

    (1976)
  • J. Flege

    Second-language speech learning: Theory, findings, and problems

  • A. Francis et al.

    Selective attention and the acquisition of new phonetic categories

    Journal of Experimental Psychology: Human Perception and Performance

    (2002)
  • M. Grammont

    Traité de phonétique

    (1965)
  • S. Guion et al.

    An investigation of current models of second language speech perception: The case of Japanese adults’ perception of English consonants

    Journal of the Acoustical Society of America

    (2000)
  • J. Harris

    Syllable structure and stress in Spanish: A nonlinear analysis

    (1983)
  • Cited by (0)

    View full text