PTLC2005 Lowie, Gilbers & Bos A close examination of L2 pronunciation: 1 A close examination of L2 pronunciation: English secondary stress by advanced Dutch learners Wander Lowie, University of Groningen Dicky Gilbers, University of Groningen Jenny Bos, University of Groningen 1 Introduction Studies on the acquisition of L2 phonology have predominantly focused on the acquisition of segments. The exception to this rule is the extensive work of John Archibald and Martha Young Scholten on the acquisition of stress in a second language. These studies, however, have concentrated on the acquisition of primary stress (see, for instance, Archibald, 1997). Very few studies have entered the intricate field of secondary stress in relation to second language acquisition. In this paper we will address the question to what extent near-native Dutch learners of English have been able to acquire nativelike use of secondary stress, being one of the subtlest of differences between these two languages. To answer this question we have conduct an experiment involving the perception and production of secondary stress. Besides the observations on whether very advanced Dutch learners of English have acquired English secondary stress patterns, we will attempt to account for their behaviour in relation to the differences between the two languages. 2. Phonology and ultimate attainment in second language acquisition Besides the question if second language learners are able to acquire a nativelike accent, especially when they start learning the second language relatively late, much attention has been given to the question why most learners have so much trouble in acquiring a second language phonological system. An influential idea to account for this observed difficulty was that of Equivalence Classification (see, e.g., Flege, Munro, & Mackay, 1995). The cause of a foreign accent, according to this view, must be sought at the level of perception. L2 sounds that are similar to L1 sounds would be the most difficult category of sounds to learn. The reason for this is that learners classify the similar sounds as identical to the L1 sound as they do not perceive a difference. Consequently, the sound will also be produced on the basis of these incorrect assumptions, leading to foreign sounding sounds. Apart from the difficulty of defining ‘new’ and ‘similar’ categories of sounds, the idea that perception is the cause of L2 production errors is appealing. Whether or not the concept of equivalence classification also applies to prosodic features of language is an empirical question. Studies that have focused on prosodic phenomena, like intonation or rhythm have mostly concentrated on languages with prosodic characteristics that are strongly deviant from English. Also much of the work done on rhythm has focused on contrasting language pairs, like English and Japanese (e.g. Nagai, 1997) and English and Spanish (e.g. Chelaflores, 1994). From the perspective of equivalence classification, it makes sense to assume that the subtle differences between two rhythmically very similar languages may cause huge problems to L2 learners. 3. Secondary stress in Dutch and English: a comparison In stress-timed languages such as English and Dutch, foot application determines the rhythmic patterns, realised PTLC2005 Lowie, Gilbers & Bos A close examination of L2 pronunciation: 2 as the division of stressed and unstressed syllables. These stress timed languages have in common that foot application of trochaic feet occurs from right to left and that primary stress occurs on one of the three rightmost syllables. It is fair to say that in general the English and Dutch prosodic structures resemble each other to a great extent. In longer words, secondary stress occurs to the left of primary stress. The actual position of secondary stress is determined by two potentially conflicting constraints. The first one can be defined as a eurhythmy-driven alignment constraint AlignLeft, which resembles the Phrasal Rule (Hayes, 1984) proposed for English, and the Hammock Rule (Van Zonneveld, 1983) proposed for Dutch rhythm patterns in longer words. AlignLeft can be defined as align the left edge of a prosodic word with a foot, which holds that the leftmost syllable is stressed. The second constraint obeys the morphological structure of the word. In derived words the rhythmic structure of the derivation maintains the rhythmic structure of the base. In other words, the rhythmic structure of related words is similar, which is equivalent to output-output correspondence (Burzio, 1998). For example, the rhythmic structure of the word pronunciation satisfies the second constraint if the second syllable is more prominent than the first one, because the second syllable in the base pronounce is more prominent than the initial one. For a particular group of words, 5-syllable words ending in –ation, the Dutch and English prosodic grammars differ slightly with respect to the position of these constraints in their ranking as depicted in the Tables below. The AlignLeft constraint seems to be more important in the Dutch ranking (cf. Visch, 1989), whereas output-output correspondence has more weight in the English ranking of constraints. Table 1a. English ranking of the two constraints: o-o correspondence and AlignLeft. constraints → pronunciation candidates ↓ pronùnciátion prònunciátion output-output correspondence base: pronóunce AlignLeft * *! Table 1b. Dutch ranking of the two constraints: o-o correspondence and AlignLeft constraints → perfectionist candidates ↓ perfèctioníst pèrfectioníst AlignLeft output-output correspondence base: perfect *! * Secondary stress is much less prominent in speech production and perception than primary stress. Whether or not secondary stress can be acoustically measured is a controversial issue. Sluijter (1995) claims that the acoustic correlates of primary stress are duration, spectral balance, pitch and intensity. Especially duration and spectral balance appear to be good primary stress cues. The results of Sluijter’s study indicate that primary stressed syllables exhibit more energy in the higher frequencies than unstressed syllables of the same melodic content and that stressed syllables are longer. Unlike these clear outcomes for primary stress, Schreuder and Gilbers (2004) found that the same cues are less reliable with respect to the identification of secondary stress. In their study on rhythmic adjustments at different speech rates, secondary stress could not be determined in the way Sluijter had done for primary stress. Instead, stressed syllables seem to appear at constant intervals of about 300 ms. For instance, the stressed second PTLC2005 Lowie, Gilbers & Bos A close examination of L2 pronunciation: 3 and fourth syllable in the Dutch word amèrikáan (‘American’) in andante tempo are at approximately the same distance from each other as the stressed first and fourth syllable in àmerikáan in allegro tempo. This suggests that prominent syllables are heard on the beats of an internal metronome, which is set to about 300 ms. The subtle difference in secondary stress between English and Dutch for the five-syllable –ation words (pronunciation) in addition to the difficulty of establishing secondary stress in acoustic studies seems to imply that this phenomenon is susceptible to equivalence classification. If learners are unable to perceive a difference between Dutch and English, they are not likely to be able to produce words in a nativelike manner in the L2. 4. Testing stress To investigate the placement of secondary stress by advanced Dutch learners of English, three related experiments were carried out: a perception test, a production test and a test in which an attempt was made to measure secondary stress acoustically. 4.1: For the perception study a recording was made of five native British English speakers reading out a list of words, consisting of target words for our study and an equal number of dummy items to obscure the goal of the study to the participants. Three trained phoneticians and ten advanced Dutch learners of English were asked to assess the stress pattern of the words. The results show that a significant majority (of about 65%) of both groups perceived secondary stress on the second syllable (pronùnciátion) rather than on the first syllable (prònunciátion) (t[11]=-3.3; p=0.008), following the English constraint ranking (OutputOutput > AlignLeft). There was no difference between the learners and the trained linguists. 4.2: In an attempt to measure the occurrence of secondary stress acoustically, native speaker recordings were made in such a way that the first, unstressed, syllable in the target word (pronùnciátion) could be matched with a secondarily stressed similar syllable from a different word (pròhibítion). The words were first edited in Cooledit and subsequently analysed in PRAAT (Boersma and Weenink, 1992-2005). The analyses concentrated on syllable duration, pitch and intensity. The results showed that in only 12 sets of syllables out of the total of 36 matched pairs could secondary stress be determined on the basis of the variables included. 4.3: Finally, a production study was carried out, in which a group of 12 advanced learners of English were asked to read out a word list consisting of target words and dummy words. The recordings were made anonymous and assessed on the position of the secondary stress by four trained linguists (Interrater reliability was 0.76). Although it seemed that in a majority of the words the English secondary stress pattern was applied, this was not significantly so, probably due to large individual differences between speakers. Therefore, no overall correctness was found in the production task. 5 Discussion and conclusion What is more important for learners of English when they are faced with English polysyllabic words that have a secondary stress, English OUTPUT-OUTPUT CORRESPONDENCE or the Dutch ALIGNLEFT constraint? This question could not be answered straightforwardly. The Dutch learners in our study appeared to be quite able to perceive the English pattern pronounced by English natives, but were not able to produce that pattern consistently. Neither did we manage to measure secondary stress acoustically. The latter observation is in agreement with what was found by Gilbers & Schreuder (2004). Since perceived secondary stress could not be PTLC2005 Lowie, Gilbers & Bos A close examination of L2 pronunciation: 4 substantiated by acoustic measurements, they claim that secondary stress is a perceptive phenomenon based on the listener’s ‘internal metronome’. However, it is not obvious how this view is compatible with the results of the current study. If secondary stress is clearly perceived in the speech of native speakers but not in the speech of advanced learners, what does that tell us about the listeners? Do they adjust to a different rhythmic pattern when they suspect the speaker is not English? Do English native speakers use a different (i.e. slower) speaking rate? At this stage we can only speculate about the explanation for this finding. Another point of interest is the role of Equivalence Classification (EC) in the acquisition of prosodic features. EC predicts that areas of the second language that are very similar to the L1 are difficult to perceive due to incorrect perception of the target language, which in turn may lead to production problems. The current study does not corroborate this position. Although we can probably speak of secondary stress as a ‘similar’ phenomenon between Dutch and English, the advanced Dutch learners in our study did not have any trouble perceiving the English pattern. Since they did show problems in producing secondary stress according to the English constraint ranking, the explanation must be sought outside perception. An additional observation that possibly hints towards a solution was that the tendency to produce noncognates more ‘correctly’ than cognates. Most problematic for the learners to produce were the English words in the test that were very similar in form and meaning to Dutch equivalents, like assìmilátion (Dutch àssimilátie) and commùnicátion (Dutch còmmunicátie). It might be that the Dutch pattern of these words has been so strongly ingrained in the minds of the Dutch speakers that they are not able to replace this pattern by the English one. This conclusion must be tentative, as the difference between cognates and noncognates did not turn out to be significant. Obviously, further research is required to fully investigate these points. 6 References Archibald, J. (1997). The acquisition of English stress by speakers of nonaccentual languages: Lexical storage versus computation of stress. Linguistics, 35, 167-181. Boersma, P & D. Weenink (1992-2005), Praat: a system for doing phonetics. http://www.praat.org. Burzio, L. (1998). Multiple correspondence. Lingua, 104, 79-109. Chelaflores, B. (1994). On the Acquisition of English Rhythm - Theoretical and Practical Issues. Iral-International Review of Applied Linguistics in Language Teaching, 32, 232242. Flege, J. E., Munro, M. J., & Mackay, I. R. A. (1995). Factors affecting strength of perceived foreign accent in a second language. Journal of the Acoustical Society of America, 97, 3125-3134. Hayes, B. (1984). The phonology of rhythm in English. Linguistic Inquiry, 15, 33-74. Nagai, K. (1997). Mora timing by British learners of Japanese. In J. Leather & A. James (Eds.), (pp. 240-248). Klagenfurt, Austra: University of Klagenfurt. Schreuder, M. J. & Gilbers, D. G. (2004). The influence of speech rate on rhythm patterns. In D.G.Gilbers, M. J. Schreuder, & N. Knevel (Eds.), On the boundary of phonology and phonetics (pp. 183-202). Groningen: University of Groningen. Sluijter, A. (1995). Phonetic correlates of stress and accent. HIL Dissertations 15 Leiden University. Van Zonneveld, R. M. (1983). Affix-Grammatica. Een Onderzoek naar Woordvorming in het Nederlands. PhD Dissertation University of Groningen. Visch, E. A. M. (1989). A Metrical Theory of Rhythmic Stress Phenomena. PhD Dissertation University of Utrecht.