TONAL MELODIES AND TONAL ALIGNMENT IN EAST NORWEGIAN Gjert Kristoffersen 1. Introduction Contrastive tonal accents (accent 1 vs. accent 2) characterize most Swedish and Norwegian dialects.1 Among the best investigated varieties from a phonetic point of view are Central Swedish (henceforth CS), especially the Stockholm variety (see e.g. Bruce, 1977; 1987; Engstrand, 1995) and East Norwegian (henceforth EN), including Oslo (see e.g. Fintoft, 1970; Haugen & Joos, 1952; Kristoffersen, 1990; 1993; 2000). These varieties can be seen as closely related phonetically, in that both encode the perceptual contrast by means of a high tone (H) realized on the stressed syllable in accent 2, followed by a low tone (L), while accent 1 instead has an L on the stressed syllable. In focused domains, the L in both accents is followed by a H, which signals focus. In compounds, the latter is realized near the final secondary stress in CS, while it is realized on the final syllable of the accentual domain in EN. Finally, in the CS variety, this H is followed by a low boundary tone (L%), while the final H in EN seems to function both as a focus marker and boundary tone. As both focus marker and boundary tone, it is boosted, while it is not so prominent in non-focus accentual domains where it only encodes the right edge boundary. When it comes to phonological interpretation of the contrast, however, two competing hypotheses exist. The first can be traced back to Haugen and Joos (1952) and to Gösta Bruce's doctoral dissertation (Bruce, 1977). Bruce assumes identical melodies for both accents, HLHL. Marking the initial H and the first L respectively as the tone that will associate with the metrically strongest syllable, i.e. the syllable that carries primary stress, then derives the accentual contrast. The contrast is usually encoded by means of asterisks associated with the H and L respectively, hence H*LHL and HL*HL. (Since only the two first tones are relevant with respect to the present discussion, I henceforth represent the contrast as H*L vs. HL*.) The initial H of the accent 1 melody will then associate with the 1 The title of the talk at Nordic Prosody IX was 'Privativity and markedness in Scandinavian tonal accents'. Thanks are due to the conference audience for useful comments, and to Jan K. Hognestad for valuable suggestions at the final stage of the writing process. unstressed syllable immediately preceding the stressed one, or delete if such a syllable is not available. Since the difference between the accents is derived from a difference in timing between the stressed syllable and the tonal melody, I shall refer to this hypothesis as the timing hypothesis. The competing hypothesis can be traced back to at least Haugen (1967), and was modelled in autosegmental terms in the late 1980s, see e.g. Kristoffersen (1990). Based on data from southern, urban East Norwegian varieties, mainly my own Arendal variety, the contrast is assumed to be a privative one encoded by the presence of an initial H in accent 2 (Kristoffersen, 2000: 241ff.), which is absent from accent 1. Again limiting the discussion to the relevant parts of the melody, the accent 2 melody can be stated as HL, while the accent 1 melody accordingly is L. Association to the rest of the prosodic structure is obtained by assuming simple left-to-right association starting from the stressed syllable. In what follows, I shall refer to this hypothesis as the privativity hypothesis. Since the two hypotheses are based on data from different varieties, it is conceivable that they both are correct, reflecting different structural properties of the two varieties. But it has later been argued that the Stockholm variety also belongs to the privativity type (Engstrand, 1995), and it has been adopted in phonological analyses of Swedish (Riad, 1998) and West Norwegian as well (Abrahamsen, 2003; Hognestad, 1997; Lorentz, 1995). Even if the question must be seen as unresolved, this suggests that the privativity hypothesis can account for both CS and EN, and that the initial H of the accent 1 melody in the timing hypothesis can be dispensed with. In this paper I shall argue that such a conclusion would be premature, however. Based on data from three East Norwegian varieties, Oslo, North Gudbrandsdal (henceforth NGbr) and Oppdal, I shall argue that at least the two latter dialects represent varieties where both accents are realized by means of a HL contour, and where the perceptual contrast is obtained by means of different timing of the H and L with respect to each other.2 The text is organized as follows: The basic hypothesis is introduced in section 2, while methodology and data will be discussed in section 3. As a basis for comparison, measurements from the Oslo dialect will be discussed briefly in section 4, while section 5 and 6 presents corresponding data from NGbr and Oppdal. Finally, some phonological implications of the findings are discussed in section 7. 2 NGbr is a rural sub-region in the northern part of the Østlandet region. Oppdal is a rural community just northeast of NGbr, across the Dovre mountain range, and about 150 km. south of Trondheim. The NGbr dialects belong to the Østlandet variety of East Norwegian, while Oppdal is one of the southernmost varieties of the Trøndelag type of East Norwegian. 2. Basic hypothesis The basic hypothesis underlying the present investigation is that if there is an initial H in the accent 1 melody, it will manifest itself on an unstressed syllable immediately preceding the stressed one linked to the L. Graphically, we can express this hypothesis as in Figure 1. V-1 V0 T1 T2 T3 V-1 V0 T1 T2 T3 Figure 1: Alternate hypotheses. Solid line = accent 1, broken line = accent 2 In the graph, V-1 and V0 represent the F0 values of two unaccented pre-stress syllables taken at the mid point of the vowels, T1 the initial H of accent 2, T2 the L common to both melodies, and T3 the final H%. At this point I assume interpolation of the accent 1 contour at T1. I shall return to this point in section 3 below. The graph to the left represents the expected result if there is an initial accent 1 H, the one to the right the expected result if there is no such tone. 3. Method and data As stated above, the basic hypothesis underlying the present investigation is that if there is an initial H in the accent 1 melody, in accordance with the hypothesis of Bruce (1977), it will manifest itself on an unstressed syllable preceding the stressed one linked to the L. The presence of such a tone can only be tested, however, if we can be sure that a H found on the preceding syllable does not have another origin. Recall that at the right edge of any accentual domain, there is a boundary H%. This means that when an accent 1 phrase follows another accent phrase, the syllable preceding the stressed accent 1 syllable will be expected to manifest a H%.3 The presence of a pre-stress, initial H belonging to the melody linked to the stressed syllable therefore cannot be tested in this environment. One possible way of overcoming this problem is to use accent phrases that are utterance initial, but preceded by at least two unstressed syllables. Since there is no evidence to my knowledge for utterance initial H% in e.g. EN, a peak on unstressed syllables in this environment immediately preceding an accent 1 L may be interpreted as evidence for such a melody initial H. In the investigation discussed in the present paper, F0 peaks and troughs have been measured as reflections of phonological tones within the target domains. A target domain is defined here as an accented, stressed syllable and a following unstressed syllable. The number of measured points in each target domain is not identical between the two accents, since accent 2 within the target domain is assumed to contain one more tone than accent 1. However, F0 has also been measured at the starting point of the stressed vowel, and in accent 1 this measure is included as 'corresponding' to the tonal peak that reflects the initial H in accent 2 words. Since no phonological tone is assumed at this point in accent 1, we will expect the F0 at this point to reflect interpolation between a (possible) tonal target associated with the pre-stress vowel and the L associated with the stressed syllable. But as will be clear in section 5 this turns out as an important diagnostic for initial H on the stressed syllable in accent 1 in Oppdal. In addition, measurements have as already noted been taken at the midpoint of the vowels of the two unstressed syllables immediately preceding the stressed syllable. The data are extracted from recordings made in 1989 and 1990 in NGbr and 1991 in Oppdal. The NGbr speakers are 6 males born between 1926 and 1952, while the Oppdal speakers are 4 males born between 1938 and 1944. (See below for information about the Oslo data.) The measurements are made on a set of sentences where each sentence consists of a fixed frame into which a disyllabic word with stress and accent on the initial syllable and with either accent 1 or 2 has been inserted. Three different frames were used. Two of them begin with Du sa ___ ('You said ___'), while the third begins with Det var ___ ('It was ___'). In all the sentence frames the expected pronunciation of the pronoun and the verb is without stress, but it turned out that several of the utterances did not meet this criterion. These have been excluded from the data set. In addition, only words with initial, voiceless obstruent and intervocalic nasal have been included in the data in order to secure maximally continuous tonal contours. A weakness of the data set is that the set of sentences recorded was not designed to elucidate the questions discussed here. As a result of this, the data set is limited and not perfectly balanced statistically across speakers and categories. 3 I define an accent phrase as the domain that includes the accented, stressed syllable at the left edge, and the syllable immediately preceding the next accented syllable (or the utterance end) at the right edge, see Kristoffersen (2000: 239f.). But across all speakers of each dialect, there is still enough data to test for significant patterns. All measurements have been entered in a database where fields for calculation of means have been added. The data were subsequently exported to an SPSS file for statistical analysis. All sentences that have been included in the data set have been controlled for possible stress on the initial syllables. Those with stress have been discarded. 4. A basis for comparison: Oslo As noted above, the evidence that has been presented till now for there not being an initial H in the accent 1 melody in EN have been indirect and scarce. However, extensive data have recently been collected from different Norwegian dialects as part of the project Norsk Tonelagstypologi (Typology of Norwegian Tonal Dialects), a project funded by the Norwegian Research Council from 2000 to 2003.4 The carrier sentence used for the data set that the following test is based on started with unstressed Jeg sa ___ ('I said ___'). The segmental make-up of the words included in the data set are more diverse than is the case for the NGbr and Oppdal data, but are in all relevant respects comparable with these, in that each consist of two unaccented followed by an accented, heavy syllable followed by at least one unstressed syllable. The speakers are three females from Oslo, all born in 1982. A total of 30 accent 1 words and 28 accent 2 words were measured. The average F0 values at each measuring points are shown in Figure 2. 300 250 A1 A2 200 150 V-1 V0 T1 T2 T3 Figure 2: Average F0 trajectories of the Oslo speakers As can be seen, there is no difference between the values at V0. This is as predicted by the privativity hypothesis, but unexpected if we go by the timing 4 For more information, see http://helmer.hit.uib.no/NTT/ hypothesis. A one-way ANOVA test performed across the four measuring points confirms the result. There is no significant difference with respect to F0 value on the pre-stress vowel (p = 0,45). Only the T1 point, where the F0 value at the beginning of the stressed vowel in accent 1 words is compared with the F0 value of the initial tonal peak in accent 2 words, shows a significant difference (p < 0,001), as would be expected. This supports the hypothesis that at least for the Oslo variety of East Norwegian, there is no initial H as part of the accent 1 melody, since we would have expected such a state of affairs to be reflected in a significant difference between the mean F0 values on the pre-stress (V0) vowel. 5. North-Gudbrandsdal We now turn to NGbr, where we shall see that the situation is different. Figure 3 shows the F0 trajectories through the measuring points for the data described in section 2. If we compare this pattern with the one we would expect if there were a H linked to the pre-stress syllable, shown in Figure 3a, we see that there is indeed a peak at V0 in accent 1. This peak could be the realization of an initial accent 1 H. A possible problem is that it is much lower than the one that represents the accent 2 peak on the stressed syllable. This difference could reflect a difference between stressed and unstressed syllables, however, given the fact that tone is used to signal prosodic prominence in Norwegian. b) a) 180 205 155 180 JB (N=4) AD (N=4) 155 130 TH (N=2) 130 105 105 80 80 V-1 V0 T1 T2 T3 SN (N=4) GS (N=2) V-1 V0 T1 T2 T3 IT (N=6) Figure 3: a) Average tonal trajectories through two unstressed syllables plus disyllabic stressed target word in NGbr. Solid line = accent 1(N=22), broken line = accent 2 (N=27). b) Individual means for accent 1 A one-way ANOVA across V-1, V0, T1 and T2 reveals that while the differences at V-1 and T2 are not significant (p = 0,470 and p = 0,280 respectively), the pvalue at V0 as well as that at T1 reaches significance (p = 0,001 and p < 0,001 respectively). This allows us to conclude with a fair amount of confidence that at least for some speakers of the NGbr variety there is a difference with respect to underlying tonal specification on the pre-stress vowel between the two accents. Of course it does not follow by necessity that the difference is caused by the presence of a H on the pre-stress syllable. But it is difficult to see any other factor that could be responsible for this difference. If it were caused by an intonational tone, e.g. an utterance initial boundary tone, we would expect this to show up on the initial syllable, and not on the second, and we would further expect it to show up irrespective of accent type. If this analysis is correct, NGbr represents a dialect that conforms to the predictions that can be deduced from the timing hypothesis, and not to those that follow from the privativity hypothesis. This means that without explicit marking of the L as the tone that associates with the stressed syllable, the accent contrast cannot be accounted for by automatic left-to-right association of tones and syllables starting from the left edge of the accentual phrase. Since the accent 2 pattern in fact can be accounted for in such a straightforward way, this identifies accent 1 as the marked member of the accentual opposition. 6. Oppdal We now turn to the Oppdal data. Let us first look at the average measurements at V0, T1, T2 and T3. (As there is no difference at V0 between the two accents, V-1 is of no interest and has been eliminated in order to avoid unnecessary detail.) They are given in Figure 4. b) a) OH 1 180 200 155 165 TV 1 130 130 LV 1 105 95 80 60 TS 1 OH 2 V0 T1 T2 T3 TS 2 TV 2 V0 T1 T2 T3 LV 2 Figure 4: a) Average F0 trajectories through one unstressed syllable plus disyllabic stressed target word in Oppdal. Solid line = accent 1(N=23), broken line = accent 2 (N=18). b) Individual means for accent 1 and 2 Also here we find no difference with respect to the pre-stress vowel, but surprisingly, in (a) we see that the F0-averages at the T1 position are nearly identical. Recall that for accent 1, T1 represents the F0 value at the beginning of the stressed vowel. With no high tone preceding the accentual L we would expect this value to be relatively low compared to the value associated with the initial H of accent 2, as we have just seen for the Oslo and NGbr varieties. In (b) we see that when the results are broken down on individual speakers, this pattern is consistently upheld, in that the accents for each individual speaker are neatly paired. A one-way ANOVA across V-1, V0, T1 and T2 reveals no significant differences: the significance levels are 0,498, 0,451, 0,581 and 0,290 respectively for the four measuring points. The accentual trajectories shown in Figure 4 suggest that for both accents the Oppdal melody is HLH, and that the initial H in both cases is realized within the stressed syllable. To the extent that there is an accentual contrast in this dialect, the contrast must therefore reside in different timing of the tones with respect to each other. That this is in fact the case is shown by the graph in Figure 5, which shows a pair of representative contours for the two accents, both starting at the onset of the stressed vowel. We see that there is no initial F0 difference. But while the accent 1 contour starts falling immediately, the accent 2 H is sustained for more than 100 ms. before it starts falling. Oppdal 200 150 F0 A1 100 A2 50 0 100 200 300 400 Time (ms.) Figure 5: Tonal accent contrast in Oppdal This means that also the Oppdal dialect must be assumed to have a HLH melody for accent 1 as well as accent 2. But it is different from NGbr in that the initial H in both cases is realized within the stressed syllable. The accentual contrast is in other words realized by the timing of the H tone with respect to the stressed syllable, which again forces different timing of the L. The latter is associated with the stressed syllable in accent 1, while it is associated with the post-stress syllable in accent 2. Note also that neither in this case is it possible to derive the contrast between the accents by unmarked left-to-right association. The accent 1 pattern would follow if we assume the mora to be the tone-bearing unit (tbu), while the accent 2 pattern would follow if the syllable is assumed to be the tbu. But accounting for the difference by assuming different tbu's for the two accents would violate the common, and in my view justified assumption that the tbu is fixed for a given dialect, be it by constraint ranking or by a constraint on representations. Finally, let us compare the Oppdal contours with corresponding contours from Oslo and NGbr, shown in Figure 6. NGbr Oslo 300 220 250 170 F0 A1 200 A2 A1 F0 120 A2 70 150 0 100 200 300 400 Time (ms.) 0 100 200 300 400 500 Time (ms.) Figure 6: Tonal accent contrast in Oslo and NGbr We see that the Oslo contrast can be associated with two properties. First, there is clear difference in F0 level at the onset of the vowel. Second, the initial H of accent 2 causes a timing difference with respect to the L. In accent 1, the L-phase is reached by the second half of the stressed syllable and continued into the second syllable, while in accent 2, it is shorter and will normally coincide with the initial phase of the second, unstressed syllable. The same points can be made with respect to the NGbr contours, except for the initial fall of the accent 1 contour, and the fact that the accent 2 H peak seems to occur somewhat later than in Oslo. This makes NGbr appear as an intermediate type between Oslo and Oppdal, where the accent 2 H is somewhat delayed compared to Oslo, perhaps caused by a H on the pre-stress syllable in accent 1, which will create a more pronounced fall to the accent 1 L in the stressed syllable in NGbr. In order to uphold the contrast, the two falls must be properly timed with respect to each other. 7. Phonological implications One of the attractive properties of the privativity hypothesis outlined in section 1 is that association between tones and tone-bearing units can be assumed as a straightforward left to right matching of the HL and L melodies with syllables, starting from the stressed syllable. But for NGbr as well as Oppdal the data suggest that the basic melody is HL both for accent 1 and accent 2. If this is correct, the simple, edge anchored linking pattern that follows from the privativity hypothesis cannot account for these data. The most important implication of the data discussed in this article is therefore that it forces us to reassess the claim that all East Norwegian tonal grammars can be accommodated by the privativity hypothesis. Since the tonal melodies are identical, left to right association from the stressed syllable without further stipulations will give the wrong result. Both for NGbr and Oppdal we need a way to ensure that the L is associated with the stressed syllable in accent 1. Clearly some kind of lexical marking is required in order to obtain the desired contrast in these dialects. Partly depending on theoretical framework, this can be done in different ways. Due to limitations of space, a formal analysis cannot be worked out here. But I would like to point out one important implication of the insight that lexical marking is necessary, already alluded to above. The accent that must be subjected to lexical marking, and therefore will appear as the marked member with respect to the other, is accent 1. This goes against what has till now been the commonly held view, namely that accent 2 due to its more complex melody is the marked member of the tonal accent opposition (Fretheim, 1976; Kristoffersen, 2000; Riad, 2003). Markedness here is measured on different scales, so the two claims are not directly compatible. But lexical marking as a way of blocking the applicability of a predictable, regular pattern can all the same be seen as a more fundamental markedness parameter than a metric based on differences with respect to size of structure. I therefore conclude that at least for the northern EN varieties discussed in this article, we can only account for the accentual contrast by marking words explicitly for accent 1 in some way or other in the lexicon. Accent 2 then becomes the unmarked default pattern. By positing accent 1 as the marked accent we bring the analysis in line with the only possible analysis of the initial stage of contrastive tonal accents in Scandinavian. The contrast developed from a situation where mono- and disyllabic words must have had different melodies. When a class of new disyllabic words established itself in the language, by suffix formation from earlier pronominal clitics signalling definite form and by epenthesis in order to eliminate marked codas, this set retained the monosyllabic melody (Oftedal, 1952; Riad, 2003) Clearly, the new disyllables with what we now refer to as accent 1, were the marked class at this stage. In Danish, words with stød historically correspond to words with accent 1 in Norwegian and Swedish. It is often assumed that stød is the marked member of the stød/non-stød opposition, see e.g. Bruce & Hermans (1999: 623). Positing accent 1 as the marked member of the accentual opposition in Norwegian and Swedish therefore brings the markedness analysis of the tonal accent and stød phenomena in the three languages in line.5 5 Arguments from a different perspective than the one assumed in this paper for accent 1 being the marked accent are given in Lahiri & al. (2004). References Abrahamsen, J. E. (2003). Ein vestnorsk intonasjonsfonologi. Unpublished Doctoral dissertation, Norwegian University of Science and Technology, Trondheim. Bruce, G. (1977). Swedish Word Accents in Sentence Perspective. Lund: CVK Gleerup. Bruce, G. (1987). How floating is focal accent? In K. Gregersen & H. Basbøll (Eds.), Nordic Prosody IV (pp. 41-91). Odense: Odense University Press. Bruce, G., & Hermans, B. (1999). Word tone in Germanic languages. In H. van der Hulst (Ed.), Word Prosodic Systems in the Languages in Europe (pp. 606658). Berlin, New York: Mouton de Gruyter. Engstrand, O. (1995). Phonetic Interpretation of the Word Accent Contrast in Swedish. Phonetica, 52, 171-179. Fintoft, K. (1970). Acoustical Analysis and Perception of Tonemes in some Norwegian Dialects. Oslo: Universitetsforlaget. Fretheim, T. (1976). On the unmarked status of Norwegian tone 1 (7). Oslo: Dept. of Linguistics, University of Oslo. Haugen, E. (1967). On the Rules of Norwegian Tonality. Language, 43, 185-202. Haugen, E., & Joos, M. (1952). Tone and Intonation in East Norwegian. Acta Philologica Scandinavica, 22, 41-64. Hognestad, J. K. (1997). Tonemer i en høytonedialekt. Oslo: Det Norske Samlaget. Kristoffersen, G. (1990). East Norwegian Prosody and the Level Stress Problem.Unpublished manuscript, Tromsø: University of Tromsø. Kristoffersen, G. (1993). An Autosegmental Analysis of East Norwegian Pitch Accent. In B. Granström & L. Nord (Eds.), Nordic Prosody VI (pp. 109-122). Stockholm: Almquist & Wiksell. Kristoffersen, G. (2000). The Phonology of Norwegian. Oxford: Oxford University Press. Lahiri, A., Wetterlin, A., & Jönsson-Steiner, E. (2004). Lexical accent in prefixed words and compounds: loans vs native vocabulary in Swedish and Norwegian. Paper presented at the Nordic Prosody IX, Lund. Lorentz, O. (1995). Tonal Prominence and Alignment. Phonology at Santa Cruz, 4, 39-56. Oftedal, M. (1952). On the origin of the Scandinavian tone distinction. Norsk Tidsskrift for Sprogvidenskap, 16, 201-225. Riad, T. (1998). Towards a Scandinavian accent typology. In W. Kehrein & R. Wiese (Eds.), Phonology and Morphology of the Germanic Languages (pp. 77-109). Tübingen: Max Niemeyer Verlag. Riad, T. (2003). Diachrony of the Scandinavian accent typology. In P. Fikkert & H. Jacobs (Eds.), Development in Prosodic Systems (Vol. 58, pp. 91-144). Berlin/New York: Mouton de Gruyter.
© Copyright 2024