Non - Native English Speakers ' Stress Patterns in Words and Sentences

Non-natiue and natiue American English speakers produced tokens of three-syllable words such as confession and four-syllable words such as confirmation, in isolation and in sentences. The acoustic-phonetic correlates of stress amplitude, fundamental frequency, and duration were measured for the first two syllables of each word. Both groups of speakers were highly uariable. Non-natiue speakers differed most from natiue speakers in their control of amplitude and duration. Introduction One of the more common reactions to English as produced by non-native speakers is that the rhythm or timing is inappropriate or incorrect. Even when nonnative English speakers have control of the segmentals, their suprasegmental patterns appear to be distinctively non-native in character. In a previous study investigating the acousticphonetic correlates of stress in non-native speech (Fokes, Bond, and Steinberg, 1983), we examined nonnative productions of stressed vs. unstressed syllables in words of varying complexity. We found that non-native speakers used appropriate acoustic correlates of stress in relatively simple words such as confess and complete. When faced with longer words such as compensation, however, non-native speakers showed considerably more difficulty, not with the acoustic correlates of stress as such, but with their appropriate distribution. The most salient characteristic of non-native productions was a tendency to over-emphasize unstressed syllables, that is to produce them with distinct vowels and with fundamental frequency, amplitude, and duration values more appropriate to stressed than to unstressed syllables. Although not directly comparable with our investigation of lexical stress, Adams and Munro (1978) report similar data concerning non-native speakers' production of sentence stress. Adams and Munro measured the fundamental frequency, amplitude, and duration of syllables in continuous text, selecting some which were judged to be stressed by a panel of listeners, and comparing these with identical unstressed syllables. Though the focus of their study was sentence stress rather than J. Fokes, Ph.D. School of Hearing and Speech Sciences Ohio University, Athens, Ohio and l.S. Bond, Ph.D. Department of Linguistics Ohio University, Athens, Ohio lexical stress, Adams and Munro noted that the "real difference between the stress production of the two groups lay not in the mechanisms they used to signal the feature (stress), but rather in their distribution of it" (p. 153), an observation suggestive of our findings concerning lexical stress. The purpose of the present study was to extend our understanding of non·native speakers' control of stress patterns by examining their productions of stressed vs. unstressed syllables in relatively long words both in isolation and in sentence context. We selected fundamental frequency, amplitude, and duration as the primary suprasegmental correlates of English stress (Lehiste, 1970). Specifically, research questions were the following; 1) How do native speakers of English employ the acoustic correlates of stress, amplitude, duration and fundamental frequency, in twoand three-syllable words? 2) Do native speakers use these correlates differently when producing the same words in isolation and in sentence context? 3) How do non-native speakers of English differ from native speakers in their use of the acoustic correlates of stress? 4) Do non-native speakers use the correlates of stress differently from native speakers when producing words in isolation and in context? 5) Do non·native speakers have similar problems with the acoustic correlates of English stress? Subjects The subjects were three American graduate students and six non-native speakers of English enrolled ina class in English pronunciation designed primarily for graduate teaching assistants. All the non-native speakers had good knowledge of English, more than adequate to pursue academic work at the university level. Each subject spoke a different language natively; Korean, Persian, Japanese, Spanish, Hausa, and Chinese (from Brunei). All the speakers were male. The number of different native languages of the subjects was advantageous, enabling us to compare English stress as produced by proficient speakers from quite different backgrounds. The acoustic correlates of stress in these languages have received varying amounts of investigation. Beckman (1982, 1985) reports that Japanese speakers use changes in fundamental frequency rather than changes in either amplitude or duration to emphasize syllables. Dauer (1983) describes Japanese as a mora-timed rather Human Communication Canada/Communication Humaine Canada, Vo!. 10, No. 3, 1986 5 than a syllable-timed language. Spanish employs fixed stress, and according to Dauer, is traditionally classified as a syllable-timed language. Although the Chinese spoken in Brunei has, to our knowledge, received no investigation, all varieties of Chinese presumably employ contour tones on syllables. Hausa is a register tone language and contrasts long and short vowels. Korean also uses contrastively long and short vowels (see Maddieson, 1984). The information about the suprasegmental structure of the native languages of our subjects is inadequate to make any clear predictions about specific difficulties they may encounter in English. Furthermore, there is some dispute about the role of transfer (interference) in the acquisition of the phonology of a second language (see Tarone, 1978). Although testing specific predictions concerning second language phonology would be valuable, we are interested in a slightly different goal: to describe one aspect of second language phonology. stress, as it is used in one language, English. If students of English as a second language experience similar difficulties with stress patterns, then these data may provide a basis for improved instructional procedures. Materials The materials were a subset of the test words used in our previous study: conclusion, completion, confession, competition, confirmation, and combination. All the words begin with the same prefix. In the first three, the prefix is reduced and the second syllable receives main stress. In the second three, on the other hand, the prefix receives secondary stress while the second syllable is reduced. Thus, the words allow a comparison of different stress patterns using similar phonology and morphology. The vowels in the stressed syllables of the threesyllable words are the tense li, u/, common in the languages of the world, and the lax I el, occurring much less frequently. The unstressed vowels are reduced. All the syllables receiving secondary stress in the four-syllable words have the same vowel; the unstressed syllables have reduced vowels. Procedure The subjects were recorded producing three tokens of each test word in isolation, in three different orders. The subjects also produced the same words three times in a sentence, answering a question. For example, when the test word was confession, the subject was asked 'Was the confession accepted?' and responded 'The confession was accepted." All the responses were of the same form, requiring the speaker to use the test word as the subject of a sentence. In this position, the test word did not receive sentence emphasis. The recordings were processed by a Voice Identification Inc. Pitch Trainer. Duration, amplitude, and fundamental frequency values of the first two syllables of each test word were obtained. The peak fundamental frequency was measured in each syllable of interest; peak amplitude was measured at the same point in the syllable. Syllable duration was measured from the release of the initial stop consonant, visible as 'noisiness' on the monitor screen, to the period of voicelessness or noisiness characterizing the intervocalic consonants Ik, p, f/. The ending of the second syllable was determined by an amplitude drop for nasals and noisiness for fricatives. Whenever the segmentation of a particular word was dubious or the values could not be recovered from the monitor screen, the measurements were made from spectrograms. For the three test words conclusion, completion, confession, the measured values allow a comparison between a reduced syllable and a syllable receiving primary stress; for the other three test words, competition, confirmation, combination, the comparison is between a syllable receiving secondary stress and a reduced syllable. Results It is possible to think of the words and their contexts as graded in complexity. The three words conclusion, completion, confession, when produced in isolation, probably represent the simplest speaking task. When serving as the subject of a simple sentence, these words present a little more difficulty. In that the stress pattern of the other three words, competition, confirmation, combination, involves secondary stress, these are probably a greater challenge for the speakers in isolation and particularly in sentence contexts. On the basis of these considerations, we would expect that the non-native speakers' acoustic-phonetic realizations of stress would be most like the American patterns for the first three words in isolation, least like the American patterns for the second three words in sentence context. Figures 1 to 3 show the relationship between stressed and unstressed syllables for the American speakers and the non-native speakers. Each data point is derived from the average values for one speaker for one word. The axes are ratios of the reduced to the stressed syllable, the y-axis representing fundamental frequency, the x-axis duration. A ratio of 1.0 means that the stressed and unstressed syllable are equal on a particular parameter; a ratio of .5 means that the unstressed syllable has half the value of the stressed syllable. Values greater than 1.0 mean that the unstressed syllable exceeds the stressed syllable on the parameter in question. Amplitude is represented as the 'tail' on each data point, the longer the 'tail', the greater the amplitude difference. When the 'tail' is above a particular data point, the amplitude relationship is reversed: the reduced syllable is produced with a greater amplitude than the stressed syllable. The data for the three American English speakers are given in Fig. 1. As can be seen from the figure, there is considerable variability in the fundamental frequency, 6 Human Communication Canada/Communication Humaine Canada, Vol. 10, No, 3, 1986 '.1 l!!!!!: .. ,,--'_10 __ 01" ..... _ • ,. .'I ........ ' •. t