Prosodic Effects on Linguopalatal Contact: Study on Initial and Final Strengthening, Lab Reports of Italian Language

Download Prosodic Effects on Linguopalatal Contact: Study on Initial and Final Strengthening and more Lab Reports Italian Language in PDF only on Docsity! Articulatory strengthening at edges of prosodic domains Cécile Fougeron Phonetics Laboratory, Department of Linguistics, University of California at Los Angeles, Los Angeles, California 90095-1543 and Institut de Phonétique, Paris Sorbonne–Nouvelle, URA 1027, France Patricia A. Keating Phonetics Laboratory, Department of Linguistics, University of California at Los Angeles, Los Angeles, California 90095-1543 ~Received 16 April 1996; accepted for publication 25 February 1997! In this paper it is shown that at the edges of prosodic domains, initial consonant and final vowels have more extreme ~less reduced! lingual articulations, which are called articulatory strengthening. Linguopalatal contact for consonants and vowels in different prosodic positions was compared, using reiterant-speech versions of sentences with a variety of phrasings read by three speakers of American English. Four prosodic domains were considered: the phonological word, the phonological ~or intermediate! phrase, the intonational phrase, and the utterance. Domain-initial consonants show more linguopalatal contact than domain-medial or domain-final consonants, at three prosodic levels. Most vowels, on the other hand, show less linguopalatal contact in domain-final syllables compared to domain-initial and domain-medial. As a result, the articulatory difference between segments is greater around a prosodic boundary, increasing the articulatory contrast between consonant and vowels, and prosodic domains are marked at both edges. Furthermore, the consonant initial strengthening is generally cumulative, i.e., the higher the prosodic domain, the more linguopalatal contact the consonant has. However, speakers differed in how many and which levels were distinguished in this way. It is suggested that this initial strengthening could provide an alternative account for previously observed supralaryngeal declination of consonants. Acoustic duration of the consonants is also affected by prosodic position, and this lengthening is cumulative like linguopalatal contact, but the two measures are only weakly correlated. © 1997 Acoustical Society of America. @S0001-4966~97!04106-4# PACS numbers: 43.70.Aj, 43.70.Fq @AL#INTRODUCTION It is by now well-established that prosody affects articu- lation. Beckman and Edwards ~1994, p. 8! define prosody as ‘‘the organizational framework that measures off chunks of speech into countable constituents of various sizes.’’ These constituents are called prosodic domains, and their organiza- tion is called prosodic structure. Prosodic structure plays an important role in the realization of the ‘‘content’’ of speech sounds ~Beckman and Edwards, 1994; Pierrehumbert and Beckman, 1988, p. 116; Fujimura, 1990b, p. 325!. Beckman and Edwards distinguish two kinds of locations within pro- sodic domains that lead to differences in the articulation of content. One location is the head, or most prominent part, of a domain. For example, a nuclear accented syllable is the head of an intermediate phrase, and its vowel can have greater duration, lingual displacement towards the target, or velocity than other vowels ~Beckman and Edwards, 1994; de Jong, 1995!. The other location is next to the boundaries of the domain, the initial and final edge positions. For example, lengthening, a temporal change, has been found to occur at the initial and final edges of prosodic domains ~e.g., Oller, 1973 for English; Byrd et al., 1997 for Tamil!. Our interest here is to add to the literature on domain edges, particularly the relatively small literature concerning spatial changes at edges. For word edges, several studies have shown that articu- lations of the tongue, lips, velum, and glottis differ in mag-3728 J. Acoust. Soc. Am. 101 (6), June 1997 0001-4966/97/101(nitude in word-initial versus non-initial position ~e.g., Fromkin, 1965; Vaissière, 1988; Krakow, 1989; Cooper, 1991; Browman and Goldstein, 1992; Farnetani and Vayra, 1996!; for example, word-initial stops in English have more linguopalatal contact ~Byrd, 1994, 1996!. For phrase and sentence edges, similar articulatory variation has been found, though the number of studies is smaller: more linguopalatal contact for sentence-initial coronal stops ~Keating, 1995!, more lip rounding for sentence-initial rounded vowels ~van Lieshout et al., 1995!. Fougeron and Keating ~1996! found less nasal airflow ~interpreted as higher velum!, and more linguopalatal contact, for French /n/ when initial in a phrase. On the other hand, Byrd et al. ~1997! found lengthening but no spatial changes at word and phrase edges in Tamil. Fi- nally, acoustic records suggest that glottal articulations also are influenced by phrasal position ~Pierrehumbert and Talkin, 1992; Jun, 1993; Dilley et al., 1996!. Another effect on articulation has also been suggested: That articulations are more extreme earlier in utterances and decline1 gradually over the course of utterances ~e.g., Vais- sière, 1986; Vayra and Fowler, 1992; Krakow et al., 1994; Hinton, 1996!. Vayra and Fowler described the articulatory variation they found for stressed Italian vowels as a ‘‘su- pralaryngeal weakening’’ and ‘‘declination of supralaryngeal gestures’’ ~p. 49!, ‘‘a weakening of the entire mechanism, respiratory, laryngeal and supralaryngeal for stressing a vowel’’ ~p. 59!. Most recently, Krakow et al. ~1994! found37286)/3728/13/$10.00 © 1997 Acoustical Society of America that the height of the velum during the English /t/ depended on its position from early to late in a sentence. The velum was highest for the earliest /t/, intermediate for a middle /t/, and lowest for the latest /t/. They described ‘‘supralaryngeal declination’’ as ‘‘a general ‘winding down’ in speech’’ ~p. 333!. The question arises, then, how supralaryngeal declina- tion, apparently a global effect of serial position in a sen- tence, is related to the local effect of being at the edge of a prosodic domain. One idea has been that they are the same thing: That domain-final articulations are reduced relative to domain-initial ones precisely because they come later in the domain ~Krakow 1989, p. 181!. Yet existing data make clear that effects at word and phrase edges cannot be ascribed to simple sentence-level declination, because those effects do not depend on serial order. For example, in Byrd ~1994!, Jun ~1993!, and Pierrehumbert and Talkin ~1992!, the position of test segments in test sentences was controlled across com- parisons. If edge effects and declination are to be related, then it must be in a more complicated way. An important finding about domain-final lengthening is relevant here. Klatt ~1975! and Wightman et al. ~1992! showed that final lengthening is found at more than one do- main level, and the lengthening is greater at the end of higher domains than of lower domains. That is, this edge effect operates hierarchically. Similarly, Jun ~1993 p. 237! pro- posed that ‘‘there is a hierarchy of strength of prosodic po- sition’’ to account for her finding that voice onset time ~VOT! of a Korean consonant is greatest when phrase-initial, next greatest when word-initial but phrase-medial, and least when word- and phrase-medial. Supralaryngeal declination could also be hierarchically nested @as is f0 declination ~Thorsen, 1985; Maeda, 1976!#, and occur not only at the sentence level, but also at the word level, and at phrasal levels in between. Under this interpretation, then, declination would not depend strictly on serial position within a sen- tence, but instead on serial position within any given pro- sodic domain. However, alternative hypotheses to declination are also potentially consistent with previous observations. Figure 1 shows a schematic of different possible patterns within a single prosodic domain, with three points highlighted in each pattern. Articulation varies along some arbitrary dimension on which lower, less extreme, values indicate articulatory reduction. We will refer to less extreme articulations as weakened ~right panels! and more extreme articulations as strengthened ~left panels!. This is the same notion of weak- ening as that used by Vayra and Fowler ~1992! and in his- torical linguistics ~Straka, 1963!. Figure 1~a! and ~b! shows progressive trends in the two directions: Progressive weak- ening, corresponding to supralaryngeal declination, and pro- gressive strengthening, a kind of reverse declination. In con- trast, the other patterns represent more localized effects at domain edges, as has been shown to be the case for domain- final lengthening ~Wightman et al., 1992; Beckman et al., 1992!. Figure 1~d! shows a pattern that we will call final weakening, a local reduction or lenition only at the end of the domain. Figure 1~e! shows the converse, what we will call initial strengthening. Initial strengthening accords with3729 J. Acoust. Soc. Am., Vol. 101, No. 6, June 1997Fujimura’s suggestion ~1990a, p. 232! that ‘‘Syllable initial position, as well as word and phrase initial position, seems to be generally characterized by more ‘forceful’ articulatory gestures ~...!.’’ Other possible local effects are given in Fig. 1~f! ~initial weakening! and ~c! ~final strengthening!, and combinations in ~g! ~V-shaped initial plus final strengthen- ing! and ~c! ~initial plus final weakening!. Any of these patterns can be hierarchically nested. If some small number of datapoints is then sampled from the whole, then it is possible to obtain a set of three declining datapoints not only from nested progressive weakening, but also from nested initial strengthening or nested final weak- ening. All that is required is that the test utterance is pro- duced with a certain kind of prosodic organization, in which the hierarchical levels decrease along with serial position in the sentence. That would be any prosodic structure in which the first datapoint is from a segment which is initial in a high prosodic domain ~e.g., an utterance!, the second segment is initial in some lower prosodic domain ~e.g., a phrase!, and the third segment is initial in an even lower prosodic domain ~e.g., a word!. There is thus a distinction between what can be observed in a set of datapoints versus the mechanisms that might underlie and produce that observation. The fact that some measure declines over an utterance does not by itself mean that speech involves a declination mechanism. There- fore our attention is focused on local effects as an alternative to declination. The present experiment was designed with four goals. The first is to determine whether the articulation of a seg- ment varies depending on its position in long sentences. The second is to determine if any such variation is due to a local strengthening or weakening at particular prosodic positions, or if it is due to a global progressive trend. To do this we compared consonants and vowels in a CV syllable in initial, medial, and final positions within each of four prosodic do- mains. The third goal is to test whether the articulatory varia- FIG. 1. A schematic of possible patterns of articulatory variation within a prosodic domain. The horizontal axis represents time; the vertical axis rep- resents an arbitrary dimension of articulatory variation, in which lower val- ues are a less extreme articulation. See text for explanation of individual panels.3729C. Fougeron and P. A. Keating: Articulatory strengthening B. Results for linguopalatal contact 1. Within-domain comparisons We tested for variation in linguopalatal contact of /n/ and /o/ within each prosodic domain in order to determine which, if any, of the mechanisms presented in Fig. 1 is at work in our data ~especially, a local effect at domain edges or a global trend!. Recall that the dependent variable, the measure of articulatory variation, is the extreme value of linguopalatal contact. For /n/, for which the extreme of con- tact is the maximum value, strengthening would correspond to an increase in contact. In contrast, for /o/, for which the extreme of contact is the minimum value, strengthening would be a decrease in contact. Weakening will show the reverse patterns. A decrease of contact for /o/ can result from a backing or lowering of the tongue, but to simplify here we will describe it as ‘‘opening.’’ a. Tests for domain initial effects: initial versus medial and final syllables. In this analysis we compare the initial syllable with the medial and final syllables in each prosodic domain to test for local effects in domain-initial position. First, U-initial segments were compared to U-medial seg- ments, i.e., all other segments ~recall that there are no U-final test segments!. Second, IP-initial segments which were not also U-initial were compared to IP-medial and IP-final seg- ments. Third, PP-initial segments which were not also IP- initial or U-initial were compared to PP-medial and PP-final segments. Finally, W-initial segments which were not also PP-initial, IP-initial, or U-initial were compared to W-medial and W-final segments. Therefore, in these analyses, ‘‘IP- initial,’’ for example, means ‘‘the highest domain in which this syllable is initial is the IP.’’ This limitation to ‘‘exclu- sive’’ coding was necessary to ensure that initial segments in some smaller domain would not have greater average contact than the medials or finals because some initials were also initial in a larger domain. With three speakers, there were thus a total of 12 comparisons. Figure 3 shows the average values for each speaker of these linguopalatal contact measures for the /n/’s and the /o/’s in initial, medial, and final positions within each pro- sodic domain. For this analysis, all data from the three sub- jects were used. Table II gives the results of statistical com- parisons by one-factor analysis of variance ~ANOVA! and Fisher’s protected least significant difference ~PLSD! post hoc pairwise comparisons, with 0.05 as the significance level for all tests. The null hypothesis is that there is no difference in contact across these positions within each domain. Consider first the /n/’s. For all three speakers, U-initial /n/’s have significantly more contact than U-medial /n/’s; IP-initial /n/’s have significantly more contact than IP-medial and IP-final-syllable /n/’s; PP-initial /n/’s have significantly more contact than PP-medial and PP-final-syllable /n/’s. More linguopalatal contact in initial position for /n/ is an initial strengthening. On the other hand, no general pattern is seen within words; only speaker 3 has more contact for W-initial /n/ than for both W-medial and W-final-syllable /n/. Speaker 1 does not show variation across word positions. For speaker 2, Word initial /n/’s have more contact than medial ones but not more than final ones.3732 J. Acoust. Soc. Am., Vol. 101, No. 6, June 1997Note that this initial strengthening holds although the prosodic coding is exclusive for domain initial /n/’s. How- ever, another possible confound in this kind of analysis could be that the higher-domain-medial /n/’s would have less con- tact simply because they would also be in medial or final syllables in the word. Nonetheless, when we compared ini- tials, medials, and finals within the PP and IP levels, taking only /n/’s which are W-initial or in operators, much the same result was found @~b! in Table II#. IP-initial /n/’s still had significantly more contact than medials for all three speakers. PP-initial /n/’s still had significantly more contact than me- dials for two of the three speakers ~speakers 1 and 2!; for speaker 3 ~the subject who has more contact W-initially than medially and finally! the direction of difference was main- tained but it was no longer significant. Thus when the word and PP levels are taken together, every speaker distinguishes initial consonants in one or the other domain. Another confounding factor could be the position of lexical stress—this could favor one possible mechanism compared to the others, especially at the word level. For speaker 1, we recorded the reiterant version of the three nu- merals in all the sentence types. When the three lexical stress FIG. 3. Maximum linguopalatal contact for /n/’s ~left! and minimum lin- guopalatal contact for /o/’s ~right! in three positions ~initial, medial, final! in each of the four prosodic domains ~utterance, intonational phrase, phono- logical phrase, word!. Speaker results are shown separately within each panel. See Table II for significance of comparisons. All data from all speak- ers are included here, coded exclusively. A more extreme articulation is more contact for /n/ and less contact for /o/.3732C. Fougeron and P. A. Keating: Articulatory strengthening TABLE II. Results of statistical comparisons for Sec. I B 1. Ficher’s PLSD post hoc comparison of percent of linguopalatal contact between initial versus medial versus final /n/’s and /o/’s at the four prosodic domains defined for the three speakers. *5p,0.05; ns5p>0.05. The columns for each domain correspond to: ~a! all data in Fig. 5, ~b! only W-initial and operators, ~c! not W-initial or PP-initial. All numerals /n/ /o/ U IP PP W U IP PP W Speaker 1 ~a! ~a! ~b! ~c! ~a! ~b! ~c! ~a! ~a! ~a! ~b! ~c! ~a! ~b! ~c! ~a! initial versus medial .* .* .* .* .* ns ,* ,* ,* ,* ,* ,* initial versus final .* .* ns ns .* .* medial versus final .* ns ns ns ns .* .* .* .* .* Speaker 2 initial versus medial .* .* .* .* .* .* ns ns ns ns ns .* initial versus final .* .* ns .* .* .* medial versus final ns ns ns ,* ,* .* .* .* .* .* Speaker 3 initial versus medial .* .* .* .* ns .* ,* ns ns .* .* ,* initial versus final .* .* .* .* .* .* medial versus final ns ,* ns ns ns .* .* .* .* .*patterns are equally represented in this way, initial strength- ening of /n/ is found at every level except the word level, suggesting that lexical stress is not the cause of the overall initial strengthening pattern. However, at the word level no distinction is found between initial, medial, and final-syllable /n/’s, suggesting that there could be an effect of lexical stress. Next consider the /o/’s, shown in the panels at the right in Fig. 3. Recall that initial strengthening, a more extreme articulation in domain initial position, would mean for /o/ less contact ~greater opening!, while initial weakening would mean more contact. Initial-syllable /o/’s ~/o/’s in the initial CV syllable of a domain! have less contact than medial /o/’s for only half of the 12 comparisons: within all levels for speaker 1, and within utterance and word for speaker 3. In two comparisons initial-syllable /o/’s have more contact than medials, and in the other comparisons they have the same contact. So for the vowel /o/, there is some initial strength- ening, but it is less prevalent than that observed for /n/, being consistent for only one speaker. These results hold whether we include only vowels in the word-initial syllable, or all3733 J. Acoust. Soc. Am., Vol. 101, No. 6, June 1997vowels, at higher prosodic levels @Table III~a! and ~b!#. b. Tests for domain-final effects: final versus initial and medial syllables. In this analysis we compare the final syl- lable with the medial and initial syllables in each prosodic domain to test for local effects in domain-final position. This test cannot be done at the utterance level, since there are no U-final test syllables. With three speakers and three levels, then, there are nine tests to be made for /n/ and for /o/. First, consider the consonants. ~Recall that final-syllable /n/’s are the /n/’s in the final CV syllable of a domain.! In Fig. 3, the difference between medial and final-syllable /n/’s can be seen to vary in size and direction across levels and speakers; sometimes medials have more contact than finals ~for ex- ample, speaker 1 in the IP!. However, again we must be careful about confounds across levels. Medials could have more contact than finals simply because more of the medials could be initial in some lower domain~s! and get strength- ened at that lower level. This is likely to be the case because domain-final syllables are almost never initial in lower do- mains. Therefore only domain-medial and domain-final- syllable /n/’s which are not initial in any lower domain ~wordTABLE III. Hierarchical prosodic levels significantly (p,0.05) distinguished by the amount of ~a! linguopala- tal contact for initial /n/; ~b! linguopalatal contact for the final vowel V1 preceding initial /n/; ~c! linguopalatal contact for the initial-syllable vowel V2 following initial /n/; ~d! C-to-V contact difference ~/n/ minus V2!; ~e! V-to-C contact difference ~/n/ minus V1!; ~f! acoustic duration of /n/; ~g! acoustic duration of final /o/ ~V1!. Results presented by speakers for the subset of data with the numeral ‘‘89.’’ Speaker 1 Speaker 2 Speaker 3 ~a! IPi.PPi.Wi5Si Ui.IPi5PPi.Wi.Si Ui5IPi.PPi5Wi.Si ~Ui5PPi and Ui5IPi! ~b! IPf,PPf,Wf5Sf IPf5PPf,Wf,Sf IPf,PPf,Sf,Wf ~c! IPi,Ui,PPi,Wi,Si Ui,IPi,Si,PPi Ui,Wi,PPi5IPi,Si ~d! IPi.Ui.PPi.Wi.Si Ui.IPi.PPi.Wi.Si Ui.IPi5PPi.Si (Wi5IPi,Wi.PPi) ~e! IPi.PPi.Wi5Si IPi5PPi.Wi.Si IPi5PPi5Si.Wi ~IPi.Wi! ~f! IPi.PPi.Wi.Si5Ui IPi5PPi.Wi.Ui.Si Ui5IPi.PPi.Wi.Si ~g! IPf5PPf.Wf5Sf IPf5PPf.Wf5Sf IPf.PPf.Sf.Wf3733C. Fougeron and P. A. Keating: Articulatory strengthening or PP! were compared @~c! in Table II, and column ~a! for the word level#. Of the nine comparisons, three give a significant difference between medial and final-syllable /n/’s: at the word and PP levels for speaker 2, and at the IP level for speaker 3, with final-syllable /n/’s having more contact than medial /n/’s, i.e., final strengthening. For the other six cases final-syllable and medial /n/’s have the same contact. For the vowels, it can be seen in Fig. 3 that /o/’s in domain-final positions generally show the least contact. Domain-final /o/’s are more open than medial /o/’s at every level for all speakers and domain-final /o/’s are usually more open than /o/’s in domain-initial syllables, with speaker 1’s IP domain the only exception. Greater opening or backing is a more extreme articulation for /o/, thus a final strengthening. This result cannot be due only to lexical stress on the final syllable of ‘‘89,’’ because in the data for speaker 1 three lexical stress patterns are included equally, and final strengthening is the same for this speaker as for the other two. c. Tests for declination within-domain and across serial position. In the previous sections we have shown that initial /n/’s are more extreme in their constriction than medial and final-syllable ones, and final /o/’s are more extreme in their opening than initial and medial ones. There is some initial strengthening of the /o/’s in the initial CV syllable, and less consistent final strengthening of the /n/’s in the final CV syllable. In this section, we compare these results to see if they form a global trend across the three positions. In Fig. 3, a declining pattern will show a progressive decrease of con- tact for /n/’s ~as they become less closed!, and a progressive increase of contact for /o/’s ~as they become more closed!. For /n/, the usual pattern is simple initial strengthening @Fig. 1~e!#, but there are three comparisons showing initial strengthening combined with final strengthening @Fig. 1~g!#. There are no cases of declination. Thus the initial strength- ening of /n/ is not part of a larger declining trend. For /o/, the relation across the three positions shows three different pat- terns. Four comparisons show a combination of initial and final strengthening @Fig. 1~g!#, and three show simple final strengthening @Fig. 1~c!#. Two comparisons show a progres- sive opening trend @Fig. 1~a!#, with final /o/’s more open than medial /o/’s and medial /o/’s more open than initial-syllable /o/’s. This progressive opening of /o/ is significant for speaker 2 at the word level, and for speaker 3 at the PP level. Thus, for these /o/’s we observed not a declination but in- stead a progressive strengthening. In sum, evidence for a within-domain declination has been found for none of the consonant comparisons; two vowel comparisons show a re- verse declination, i.e., progressive strengthening. We also tested for a sentence-level global trend that would depend only on serial position of a segment in the sentence. We tested whether the amount of linguopalatal contact varies linearly, either over the whole sentence or within smaller domains. This was done by testing for corre- lations between linguopalatal contact and linear position in the sentence, regressing serial position of the consonants ~from syllable 1 to syllable 15! against their amount of con- tact, and using all data from all speakers. In the previous analysis we averaged all domain-medial syllables and com-3734 J. Acoust. Soc. Am., Vol. 101, No. 6, June 1997pared them as a group to averaged domain-initial ones and domain-final ones; here we code all syllables by serial posi- tion so each could show its contribution to a potential pro- gressive trend. We found no such trends for /n/’s or /o/’s ~r2<0.01) for all comparisons: over /n/’s in all syllables; stressed syllables; W-initial syllables; PP-initial syllables; IP-initial syllables; over /n/’s only in medial and final syl- lables, taking out the strengthened W-initial, PP-initial, IP- initial, and U-initial syllables; over all /o/’s or ~following Vayra and Fowler, 1992! stressed /o/’s only!. Thus we can say that in our data there is neither overall declination nor overall progressive strengthening for either consonants or vowels. 2. Hierarchical level of the domain boundary (across- domain comparisons) a. Effects on linguopalatal contact. The next analysis focuses on the hierarchical nature of the strengthenings found above. For example, we ask whether the initial strengthening found for /n/ in different prosodic domains is cumulative, whether contact for initial consonants is greater when they are initial in higher prosodic domains than when they are initial only in lower prosodic domains. The com- parisons were made by ANOVA followed by Fischer LPSD post hoc comparisons, with 0.05 as the significance level, for all pairings of domain initial /n/’s: W-initial /n/’s which are not also initial in PP, IP, or U; PP-initial /n/’s which are not also initial in IP or U; IP-initial /n/’s which are not also initial in U; and U-initial /n/’s. In order to compare W-initial consonants with ones initial in a lower domain, we included in the comparison syll-initial consonants that are not also W-initial or W-final ~i.e., W-medial consonants!. Similar comparisons are made with domain-final vowels and vowels in domain-initial syllables. Figure 4~a! shows the average maximum contact for ini- tial /n/’s according to the consonants’ highest domain, for just the sentences with ‘‘89’’ ~the numeral for which all the speakers produced all the sentence types! for the individual speakers. The general tendency is that the contact is related to the hierarchical level of the domain boundary: Higher lev- els show more contact. However, as can be seen in the figure, speakers vary. Not all levels are reliably distinguished, and although three or four levels are distinguished, the speakers differ in which levels those are. Table III~a! summarizes the significant differences for each speaker. In speaker 1’s data, IP-initial, PP-initial, and W-initial/syll-initial /n/’s are dis- tinct, but U-initial is not different from either IP-initial or PP-initial, and W-initial is not different from syll-initial ~W- medial! ~as already seen in the previous section!. Speaker 2 distinguishes U-initial, IP-initial/PP-initial, W-initial, and syll-initial, but IP-initial is not different from PP-initial. Speaker 3 distinguishes IP-initial/U-initial, W-initial/PP- initial, and syll-initial, but U-initial and IP-initial are not dif- ferent from each other, nor are PP-initial and W-initial ~as already seen in the previous section!. In sum, syllable is dis- tinguished from word ~2 speakers!, word is distinguished from PP ~2 speakers!, and PP is distinguished from a higher domain ~3 speakers!, but whether that higher domain is the IP or the utterance is variable.3734C. Fougeron and P. A. Keating: Articulatory strengthening depends on whether initial strengthening can distinguish three prosodic levels. Initial strengthening would give a pat- tern looking like declination if the consonants compared are initial in these three prosodic domains and if the hierarchical levels of these domains decreases along with their serial po- sition in the sentence. To see this, consider again Fig. 2, our sample data, for example the top utterance. It is possible to pick three datapoints for /n/ which show declination: for ex- ample, the first /no/ ~which is U-initial and IP-initial!, the eighth /no/, corresponding to the second ‘‘plus’’ ~which is PP-initial!, and the 13th /no/, corresponding to the beginning of the last numeral ~which is W-initial!. This apparent decli- nation is only the consequence of three different degrees of strengthening at the word, PP, and IP levels. If only these points had been considered, we would have concluded that there is declination. Yet all of our systematic comparisons testing for declination yielded none. B. What is the nature of this strengthening? The terms ‘‘strengthening’’ and ‘‘weakening,’’ as well as the variants ‘‘fortition’’ and ‘‘lenition,’’ are often used to characterize segment variation or historical changes ~Hock, 1992!, but it is seldom that those terms are phonetically or articulatorily defined, or that the mechanism leading to the variation is explained. Our results lead us to think that the variation we observe is the result of a general phenomenon in speech, which we call articulatory strengthening at prosodic-domain edges. We have considered this strengthen- ing to mean more extreme articulation, that is, spatial varia- tion. Strengthening may also involve greater lengthening, that is, temporal variation. Our observations of linguopalatal contact variation are only the result of this strengthening and are not sufficient to establish its nature. Here we will propose and discuss some possible mechanisms that may induce the more extreme articulations of strengthened segments. ~1! Increased duration. In general, shorter durations of- ten ~though not necessarily! lead to articulatory undershoot ~Lindblom, 1963; Moon and Lindblom, 1994!. Conversely, stronger segments could have more extreme articulations be- cause they are longer, and thus have time to reach their tar- gets. However, in our data the spatial variation observed is not strongly correlated with the temporal variation. ~2! Increased distance between segments. Recall that there is greater difference in linguopalatal contact between /n/ and /o/ at higher prosodic boundaries, for V-to-C and especially for C-to-V. Possibly, after an open final vowel there could be overshoot of the following lingual target. This was suggested for the jaw in /ata/, /asa/, and /ada/ ~though not /ana/! sequences by Keating et al. ~1994!, or, since movement velocity is usually proportional to displacement, the larger displacements from an open final vowel to the following initial consonant would involve higher velocities. Higher velocity would result in a greater impact of the tongue against the palate at closure ~A. Löfqvist, personal communication!. In both cases ~overshoot or higher velocity! more compression of the tongue tissue, and therefore greater contact, would result. But in both cases if it is /n/’s distance from the preceding vowel that leads to /n/’s greater contact, then we should see a negative correlation between the con-3737 J. Acoust. Soc. Am., Vol. 101, No. 6, June 1997tact for V1 and C. This is so for speaker 1, C has the most contact just when V1 has the least. And speaker 1 is the one subject who has consistent initial strengthening of /o/, V2 has the least contact just when C has the most. So for this subject, the spatial distance between successive oral targets could be an important factor in initial strengthenings. Whether the relevant distances, movements, and velocities are those of the tongue, the jaw, or both, cannot be addressed with EPG. ~3! Increased coarticulation. Strengthening could in- crease the overlap of /n/ with surrounding vowels, in which case contact at the back of the palate for /o/ could occur during /n/, increasing the total contact measured for /n/. In this case there should be some correlation between the con- tacts for adjacent /n/’s and /o/’s, such that when one has more contact the other also has more contact. For one speaker ~speaker 2! there is a weak correlation of this kind (r250.2); for the other speakers the correlations were either near zero, or showed less coarticulation. In general these correlations and the increased displacement between C and V’s at higher prosodic boundaries do not support this hy- pothesis. Furthermore, this hypothesis would have nothing to say about the decreased contact for strengthened /o/. ~4! Greater coarticulatory resistance. Conversely, stron- ger segments could resist coarticulatory undershoot because they resist blending with overlapping gestures ~Fowler and Saltzman 1993, p. 182!. Strengthening of /n/ would involve more contact because the tongue blade is not pulled away from its constriction target by /o/’s tongue body articulation, and vice versa for /o/. This hypothesis is supported by the increased V-to-C and ~especially! C-to-V displacements at higher prosodic levels. ~5! Increased effort or energy. Articulatory strengthen- ing could also result from a greater overall effort in speech that would also affect the pulmonary and laryngeal systems, as has been proposed for stress ~Ladefoged, 1967; Sluijter et al., 1997!. Variations observed in supralaryngeal articula- tions would only be an indirect effect of this overall energy increase ~Öhman, 1967; Vayra and Fowler, 1992!. Another possibility could be that initial strengthening is the result of a localized increase in supralaryngeal articulatory effort. Fujimura ~1990a! suggested that prosodic-domain initial consonants are characterized by more ‘‘forceful’’ articula- tory gestures. Variation of articulatory effort was also sug- gested by Straka ~1963 p. 91!, who defined articulatory en- ergy in terms of the force of contraction of the muscles primarily involved in the articulation of the segment, specifi- cally excluding respiratory and laryngeal systems. Straka found that in more ‘‘forceful’’ ~‘‘renforcée’’! pronunciation there was greater linguopalatal contact for consonants and less contact for all vowels, and as a consequence an in- creased difference in openness between successive segments. This is what we found at prosodic domain edges. In conclusion, there are a number of possible mecha- nisms that might result in the strengthening we have ob- served, but resolving this question would require much fur- ther work.3737C. Fougeron and P. A. Keating: Articulatory strengthening C. Enhancement and the listener The mechanisms discussed above present strengthening as either the automatic, unplanned consequence of some other aspect of speech production, or something learned as part of the language. In either case it might be useful linguis- tically. We can think of three ways in which initial strength- ening could benefit a listener. Two of these have to do with prosodic parsing. First, strengthening could help with seg- mentation of the signal into words and higher domains. Re- call that in Sec. I B 2 b it was shown that there is some enhancement of V-to-C and C-to-V linguopalatal contact dif- ferences at prosodic boundaries. Figure 5 combines these two aspects into a single scheme for a CV]CV sequence, where ] is some prosodic boundary. This schematic shows that the articulatory contrast between the consonant after the boundary and its surrounding vowels is enhanced because they are more extreme in opposite directions. This articula- tory enhancement of the contrasts within the sequence may contribute to marking the prosodic boundary even more clearly than do the vowel or consonant alone. A similar en- hancement of CV contrasts at domain-initial boundaries in Italian has been discussed by Farnetani and Vayra ~1996, p. 12!: ‘‘Also boundaries are signalled by an increase in CV contrast: initial boundaries are marked by a strengthening of consonant closure and by an increase in vowel posteriority ~...!.’’ Second, the degree of strengthening could possibly tell the listener about the strength of the prosodic boundary, similar to the way that Wightman et al. ~1992! suggest that listeners could use degree of final lengthening. A listener would know that when an initial consonant is more than minimally strengthened, the boundary ~or break! must be stronger. In particular, IP boundaries could be distinguished from word boundaries in this way. Our results would not support a stronger claim, that listeners might judge the abso- lute level of any new domain from the degree of strengthen- ing. There is simply too much interspeaker variation. The third way in which initial strengthening could ben- efit a listener concerns lexical access. If initial strengthening enhances the segment-specific articulations of consonants and vowels, then it could enhance cues that aid in identifying each segment. de Jong ~1995! proposes that stress involves a local hyperarticulation that makes each segment more differ- ent from all other segments of the language, so that lexical FIG. 5. Schematic summarizing linguopalatal contact for segments spanning a boundary. Dashed line shows difference between final-syllable /n/ and final /o/ in a domain; bold line shows difference between final /o/ and initial /n/; dotted line shows difference between initial /n/ and initial-syllable /o/.3738 J. Acoust. Soc. Am., Vol. 101, No. 6, June 1997contrasts are more distinctive. Along the same lines, en- hanced accessibility of segmental information in domain- initial positions would be particularly helpful, particularly for word-initial segments which are important in word rec- ognition ~Cole et al., 1978; Hawkins and Cutler, 1988! and at domain beginnings where there is less top-down ~e.g., syn- tactic and semantic! information available. Of course the linguistic function of initial strengthening presupposes that strengthening has one or more acoustic/ auditory correlates. We found that the variation in lin- guopalatal contact for /n/ and /o/ is accompanied by variation in acoustic duration, though the two measures are not strongly correlated. The acoustic duration differences would be potentially available to listeners. It remains to be seen whether the linguopalatal contact differences have any other associated acoustic properties, and whether these can be heard by listeners. D. Prosody and articulation Our results underline a point made by a few other re- searchers: the importance of understanding, controlling, and reporting the prosody of speech materials in articulation ex- periments. For the individual experimenter, unsought varia- tion in prosody is a potential confound both within and across speakers, as our own experiment shows. It can also make comparisons across studies difficult or impossible, as researchers have always known. Yet, at the same time, awareness of prosodic differences between sentences can turn apparent random variation into predictable, lawful regu- larities of speech production, as we hope to have shown here. This is not to say that prosody is easy to control, or that experiments on prosodic effects are easy to design. It can be especially difficult to find sequences of real words that can occur across a variety of prosodic boundaries and that con- tain segments appropriate to a given method of articulatory data collection ~in our case, lingual consonants!. Reiterant speech finesses this difficulty, but some subjects may not be able to produce it fluently ~Larkey, 1983! ~thought this was not a problem in the present study!. It may also be the case that reiterant speech induces somewhat exaggerated rhyth- mic alternations. These in turn may enhance the prosodic phrasing, and thereby its manifestation in articulation. The present experiment does not address this point, and our re- sults need to be confirmed with experiments using real words ~see Fougeron and Keating, 1996 for French!. Another difficulty in such experiments is that the phras- ings of the test sentences are unclear unless the utterances are prosodically analyzed. The use of orthographic devices such as punctuation ~e.g., commas! or parentheses in the test ma- terials does not guarantee that subjects will produce any par- ticular phrasing. In our study there was variation both within and across speakers, requiring post hoc prosodic transcrip- tion to determine the actual phrasing of each token. E. Conclusion Most previous research on prosodic demarcation has fo- cused on the ends of prosodic domains. Our results add to the much smaller literature on beginnings of domains. At3738C. Fougeron and P. A. Keating: Articulatory strengthening least some tongue blade and body articulations are more ex- treme in domain-initial positions. We have shown that posi- tion with respect to prosodic boundaries affects articulatory constriction, something that would seem to be an inherent property of a sound. ACKNOWLEDGMENTS This work was supported by an MRT allocation to the DEA de Phonétique de Paris to the first author and by NSF Grant No. SBR-9511118 to the second author. Earlier ver- sions were those presented at the Spring 1995 meeting of the Society in Washington DC, and appearing in UCLA Work- ing Papers in Phonetics 92 ~1996!. We wish to thank the subjects ~Kimberly Thomas and Barbara Blankenship! for participating; Sun-Ah Jun and Sabrina Cooper for help with data analysis; Janet Pierrehumbert for providing information about the sentences in Pierrehumbert and Talkin ~1992!; and Dani Byrd, Edda Farnetani, Bruce Hayes, Sun-Ah Jun, Rena Krakow, Stefanie Shattuck-Hufnagel, Jacqueline Vaissière, John Westbury, Jie Zhang, Editor Anders Löfqvist, Janet Fletcher, and two anonymous reviewers for helpful com- ments. 1The term ‘‘declination’’ originally comes from the study of intonation, where it refers to a downtrend in fundamental frequency ( f0) @see Ladd ~1984! or ’t Hart et al. ~1990! for reviews#. Subglottal pressure also de- clines over an utterance, and some of the downtrend in f0 has been attrib- uted to the downtrend in subglottal pressure ~Gelfer et al., 1987!. 2The first syllable of ‘‘70’’ and the second syllable of ‘‘100’’ were stressed for all three speakers, and other syllables in those numerals were un- stressed. For two speakers, the last syllable of ‘‘89’’ was stressed and the other syllables unstressed, but for one speaker ~speaker 3! both the first and last syllables of ‘‘89’’ were stressed and only the middle syllable was unstressed. Abacus Concepts ~1992!. StatView ~Abacus Concepts, Inc., Berkeley, CA!. Beckman, M., and Edwards, J. ~1994!. ‘‘Articulatory evidence for differen- tiating stress categories,’’ in Phonological Structure and Phonetic Form: Papers in Laboratory Phonology III, edited by P. A. Keating ~Cambridge U.P., Cambridge, England!, Chap. 2, pp. 7–33. Beckman, M., Edwards, J., and Fletcher, J. ~1992!. ‘‘Prosodic structure and tempo in a sonority model of articulatory dynamics,’’ in Papers in Labo- ratory Phonology II: Gesture, Segment, Prosody, edited by G. Docherty and D. R. Ladd ~Cambridge U.P., Cambridge, England!, pp. 68–86. Beckman, M., and Pierrehumbert, J. ~1986!. ‘‘Intonational structure in En- glish and Japanese,’’ Phonology ~Yearbook! 3, 255–310. Browman, C., and Goldstein, L. ~1992!. ‘‘Articulatory phonology: an over- view,’’ Phonetica 49, 155–180. Byrd, D. ~1996!. ‘‘Influences on articulatory timing in consonant se- quences,’’ J. Phon. 24, 209–244. Byrd, D. M. ~1994!. ‘‘Articulatory timing in English consonant sequences,’’ Ph.D. dissertation, UCLA, distributed as UCLA Working Papers in Pho- netics 86, 1–196. Byrd, D., Flemming, E., Mueller, C. A., and Tan, C. C. ~1995!. ‘‘Using regions and indices in EPG data reduction,’’J. Speech Hear. Res. 38, 821– 827. Byrd, D., Narayanan, S., Kaun, A., and Saltzman, E. ~1997!. ‘‘Phrasal sig- natures in articulation,’’ to appear in Proceedings of Laboratory Phonol- ogy V ~Cambridge U.P., Cambridge, England!. Cole, R., Jakimik, J., and Cooper, W. ~1978!. ‘‘Perceptibility of phonetic features in fluent speech,’’ J. Acoust. Soc. Am. 64, 44–56. Cooper, A. ~1991!. ‘‘Laryngeal and oral gestures in English /p,t,k/,’’ Pro- ceedings of the XIIth International Congress of Phonetic Sciences 2, Uni- versity of Provence ~Aix-en-Provence, France!, pp. 50–53. de Jong, K. ~1995!. ‘‘The supraglottal articulation of prominence in English: Linguistic stress as localized hyperarticulation,’’ J. Acoust. Soc. Am. 97, 491–504.3739 J. Acoust. Soc. Am., Vol. 101, No. 6, June 1997Dilley, L., Shattuck-Hufnagel, S., and Ostendorf, M. ~1996!. ‘‘Glottalization of word-initial vowels as a function of prosodic structure,’’ J. Phonetics 24, 423–444. Farnetani, E., and Vayra, M. ~1996!. ‘‘The role of prosody in the shaping of articulation in Italian CV syllables,’’ in Proceedings of the 1st ESCA Tutorial and Research Workshop on Speech Production Modeling and 4th Speech Production Seminar, Autrans, France, edited by P. Perrier, F. Bus- tarret, and R. Laboissière ~Institut National Polytechnique de Grenoble, France!. Fougeron, C., and Keating, P. ~1996!. ‘‘Variations in velic and lingual ar- ticulation depending on prosodic position: Results for two French speak- ers,’’ UCLA Working Papers in Phonetics 92, 88–96 ~Also available elec- tronically in the European Student Journal of Language and Speech at http://web-sls.essex.ac.uk/web-sls/!. Fowler, C. A., and Saltzman, E. ~1993!. ‘‘Coordination and coarticulation in speech production,’’ Lang. Speech 36, 171–195. Fromkin, V. ~1965!. ‘‘Some phonetic specifications of linguistic units: an electromyographic investigation,’’ UCLA Working Papers in Phonetics 3, 1–184. Fujimura, O. ~1990a!. ‘‘Methods and goals of speech production research,’’ Lang. Speech 33, 195–258. Fujimura, O. ~1990b!. ‘‘Articulatory perspectives of speech organization,’’ in Speech Production and Speech Modelling, edited by W. J. Hardcastle and A. Marchal ~Kluwer, Dordrecht!, pp. 323–342. Gelfer, C. E., Harris, K. S., and Baer, T. ~1987!. ‘‘Controlled variables in sentence intonation,’’ in Laryngeal Function in Phonation and Respira- tion, edited by T. Baer, C. Sasaki, and K. S. Harris ~Little, Brown, Boston, MA!, pp. 422–435. Hawkins, J., and Cutler, A. ~1988!. ‘‘Psycholinguistic factors in morpho- logical asymmetry,’’ in Explaining Language Universals, edited by J. Hawkins ~Blackwells, Oxford!, pp. 281–317. Hinton, V. A. ~1996!. ‘‘Interlabial pressure during production of bilabial phones,’’ J. Phonetics 24, 337–349. Hock, H. H. ~1992!. ‘‘Initial strengthening,’’ in Phonologica 1988, edited by W. U. Dressler et al. ~Cambridge U.P., Cambridge, England!, pp. 101– 110. Jun, S. ~1993!. ‘‘The phonetics and phonology of Korean prosody,’’ Ph.D. dissertation, Ohio State University. Keating, P. A. ~1995!. ‘‘Effect of prosodic position on /t,d/ tongue/palate contact,’’ poster presented at the XIIIth International Congress of Phonetic Sciences, Stockholm, 1995 ~unpublished!. Keating, P. A., Lindblom, B., Lubker, J., and Kreiman, J. ~1994!. ‘‘Vari- ability in jaw height for segments in English and Swedish VCVs,’’ J. Phonetics 22, 407–422. Klatt, D. H. ~1975!. Vowel lengthening is syntactically determined in a connected discourse,’’ J. Phonetics 3, 129–140. Krakow, R. A. ~1989!. ‘‘The Articulatory Organization of Syllables: A ki- nematic analysis of labial and velic gestures,’’ Ph.D. dissertation, Yale University. Krakow, R. A., Bell-Berti, F., and Wang, Q. E. ~1994!. ‘‘Supralaryngeal declination: evidence from the velum,’’ in Producing Speech: A Festschrift for Katherine Safford Harris, edited by F. Bell-Berti and L. Raphael ~AIP, Woodbury, NY!, Chap. 23, pp. 333–353. Ladd, D. R. ~1984!. ‘‘Declination: A review and some hypotheses,’’ Pho- nology ~Yearbook! 1, 53–74. Ladefoged, P. ~1967!. Three Areas of Experimental Phonetics ~Oxford U.P., London!. Larkey, L. ~1983!. ‘‘Reiterant speech: An acoustic and perceptual valida- tion,’’ J. Acoust. Soc. Am. 73, 1337–1345. Lindblom, B. ~1963!. ‘‘Spectrographic study of vowel reduction,’’ J. Acoust. Soc. Am. 35, 1773–1781. Löfqvist, A. ~1996!. Personal communication. Maeda, S. ~1976!. ‘‘A characterization of American English Intonation,’’ Ph.D. dissertation, MIT. Moon, S. J., and Lindblom, B. ~1994!. ‘‘Interaction between duration, con- text, and speaking style in English stressed vowels,’’ J. Acoust. Soc. Am. 96, 40–55. Öhman, S. ~1967!. ‘‘Word and sentence intonation: a quantitative model,’’ STLQ Progr. Status Rep. Stockholm 2/3, 20–54. Oller, D. K. ~1973!. ‘‘The effect of position in utterance on speech segment duration in English,’’ J. Acoust. Soc. Am. 54, 1235–1247. Pierrehumbert, J., and Beckman, M. ~1988!. Japanese Tone Structure ~MIT, Cambridge, MA!. Pierrehumbert, J., and Talkin, D. ~1992!. ‘‘Lenition of /h/ and glottal stop,’’3739C. Fougeron and P. A. Keating: Articulatory strengthening