Oceanic Linguistics, Volume 45, no. 1 (June 2006) © by University of Hawai‘i Press. All rights reserved.
The Prosodic Realization of Negation in
Saisiyat and English
Wen-yu Chiang, I Chang-Liao, and Fang-mei Chiang
graduate institute of linguistics, national taiwan universityThis study investigates the prosodic realization of negation in Saisiyat, an endangered aboriginal Austronesian language of Taiwan, and compares the prosodic properties of its af²rmative and negative sentences with those of Brit-ish EnglBrit-ish. In order to test Yaeger-Dror’s “Cognitive Prominence Principle,” according to which cognitively prominent items (such as negators) should be prosodically marked, we measure the F0 peak, the intensity peak, and duration
of lexical items appearing in af²rmative and negative sentences. Our results indicate that sentential subjects are the most acoustically prominent items with respect to F0 height and intensity in Saisiyat negative sentences, whereas the
negator itself is the most acoustically prominent item with respect to F0 in an
English sentence. In addition, the presence of a negator does not signi²cantly change the prosodic parameters of contiguous words in Saisiyat. English, in contrast, exhibits relatively large-scale prosodic differences in both F0 and
intensitybetween af²rmative and negative sentences. This paper suggests that the following typological features can account for the differences observed between Saisiyat and English: (1) the relationship between prosodic promi-nence and syntactic subjects in Saisiyat, (2) transparency of the negation sys-tem in Saisiyat, and (3) the relationship between prosodic prominence and semantically de²ned focus in English.
1. INTRODUCTION.1 Cross-linguistically, negation can be realized by means of
syntactic marking, prosodic marking, or a combination of the two, depending on the prosodic characteristics of the language in question. Syntactically, a language can use a single negator to express negation, such as not in English, or it can use various negators for different syntactic structures, such as those found in the Austronesian languages of Taiwan. Negation can also be marked prosodically. For example, Yaeger-Dror (2002) found that the F0 (fundamental frequency) of English “not” and French “pas” are higher than the F0 of surrounding words. The current literature suggests that the combi-nation of prosodic and syntactic marking of negation is relatively common; the use of
1. An earlier version of this paper was presented at the 11th Annual Meeting of the Austronesian Formal Linguistics Association (AFLA) and appeared in Proceedings of AFLA 11, ZAS (Ber-lin, Germany), 59–72. Funding for this study was provided by the National Science Council of Taiwan (grant number NSC92-2411-H-002-079). We would like to express our thanks to the three informants, Oebaj a Oemaw, Pawaj a Tahis, and Terry for their participation and enthusi-asm. Special thanks go to Tanya Visceglia, Shuanfan Huang, and to anonymous reviewers for their valuable comments. All remaining errors of commission or omission are our own.
only prosodic marking to realize negation, in contrast, is relatively rare (for a detailed discussion, see Remijsen 2003).
The question of whether a negator is invariably more prominent acoustically than its surrounding words, however, remains unresolved. Semantically, negators generally contribute focused or new information to a sentence; for this reason, they assume “focal prominence” (Yaeger-Dror 2002:1496). Both theoretical (Horn 2001, Kadman 2001) and empirical studies (Hirschberg 1990, 1993)2 support the claim that negators contribute salient information to the discourse; it is thus likely that the information they convey is also cognitively prominent. According to the “Cognitive Prominence Princi-ple” (Yaeger-Dror 2002), acoustic prominence enhances discourse participants’ atten-tion to focused items, which maximizes the effectiveness of communicaatten-tion. Based on Bolinger’s (1978) claim that prosodic marking of focused information is a linguistic universal, cognitively prominent items should therefore also be prosodically salient. Cutler, Dahan, and Donselaar (1997) proposed a “Cognitive Prominence Corollary,” according to which “a prosodically nonprominent token of a highly signi²cant word is quite unlikely” (as cited in Yaeger-Dror 2002:1497).
Despite the wide range of cross-linguistic variation in the prosodic realization of negators (see Yaeger-Dror 2002 for a comprehensive review), acoustic evidence has been found to support the “Cognitive Prominence Principle.” In O’Shaughnessy and Allen (1983), measurement of the F0 values produced by speakers within negative sen-tences determined that the F0 of negators was usually higher than that of contiguous lexical items. However, they found the F0 of negators to be raised only in isolated utter-ances. When negators were uttered within larger units of discourse such as paragraphs, F0 raising was found only rarely.3 Other studies, such as that of English not (Hirschberg 1990) and that of French pas (Morel 1995) found negators to be consistently promi-nent with respect to F0 in read speech.
Bolinger (1978) provided an alternative analysis of the prosodic characteristics of negators. Although he agreed that new information is usually prosodically marked, he claimed that cross-linguistic prosodic realization of negative sentences tends to favor lower, rather than higher F0. Yaeger-Dror (1985) interpreted this observation to mean that the F0 on negative particles is made prominent by being produced lower in the speaker’s F0 range. In the same study, however, Yaeger-Dror claimed that F0 lowering of negators has never been found in research on English prosody. Instead, the F0 level of English negators was found to be either higher than or even with the other constituents.
In addition to the controversy over whether negators are actually prosodically prominent, it also has yet to be determined whether the prosodic prominence that has
2. We cited these works from Yaeger-Dror (2002).
3. According to Yaeger-Dror (1985, 2002), the unstable behavior of negators stems from the con³ict between the “Cognitive Prominence Principle” and the “Social Agreement Principle.” On the one hand, a negator, due to its sentential prominence, must be prosodically marked in order to comply with the Cognitive Prominence Principle. On the other hand, the Social Agree-ment Principle discourages emphasis on any objection to a conversation partner’s previous assertion, which would effectively eliminate any prosodic prominence assigned to negators. Yaeger-Dror’s research suggests that the choice to assign prosodic prominence to negators may be sensitive to the dynamics of culture and discourse. Because our experimental materials con-tain only short sentences, we do not test the “Social Agreement Principle” in this study.
been attributed to English negative sentences can be found across a range of language types. The current study focuses on the prosodic realization of negators in Saisiyat,4 an endangered aboriginal Austronesian language of Taiwan, and compares its prosodic realization of negation with that of English.5 Although this research focuses on acous-tic measurement of the negators themselves, it also measures the F0 and intensity of the lexical items surrounding the negator. The prosodic characteristics of negative Saisiyat sentences are compared with those of af²rmative sentences. In addition, differences between negative and af²rmative sentences are compared between Saisiyat and English. Acoustic parameters to be measured include F0 peak, intensity peak, and duration. These results provide data for typological comparison of the prosodic realiza-tion of negarealiza-tion, as well as informarealiza-tion on the intonarealiza-tional structure of Saisiyat.
The signi²cance of choosing Saisiyat for comparison with English lies in its morphosyntactic structure; agreement between the verb and its arguments depends on whether a sentence is agent-focused (af) or nonagent focused (naf). Focus markers are obligatory in Saisiyat; this requirement is common among the Austronesian lan-guages of Taiwan.6 Focus markers are morphologically af²xed to the verb, and Saisiyat licenses only one focus marker per verb. Verbal af²xation can be used to mark agreement between the verb and any constituent, including agents, patients, locations, and other arguments. Yeh (2000) divided focus markers into two groups and identi²ed four different kinds of focus markers within those groups; these were classi²ed accord-ing to the relationship between the verb and the semantic roles of its arguments. This is illustrated in table 1:7
4. Saisiyat is an endangered language in Taiwan with about 7,000 speakers. It belongs to the Austronesian language family and has two major dialects—Northern Saisiyat (the Taai dia-lect) and Southern Saisiyat (the Tungho diadia-lect). The speaker population of these two dialects of Saisiyat lives in central Taiwan. Northern Saisiyat is spoken in Hsinchu County and South-ern Saisiyat is spoken in Miaoli County.
5. The English-speaking informant in this study is a male speaker of UK English.
6. The traditional view of “focus” in Austronesian languages is that “the focused NP in Austro-nesian languages not only functions as syntactic pivot for relativization and verb serialization, but also has the pragmatic effect of highlighting it as the center of attention in a clause”(Huang 2002:666). This position has been challenged in recent literature (e.g., see Huang 2000 for a detailed discussion). Moreover, some researchers refer to this kind of focus system as a voice system or a trigger system (e.g., Shibatani 1988, Rau 1992, Wouk 1996). According to Rau’s (1992:36) analysis, verbs in the Austronesian languages of Taiwan “usu-ally contain a root or base and one or more af²xes. The base provides the meaning of the verb, whereas the primary af²xes show the perspective of the sentence and designate the role rela-tionship of the participant to the event in question.”
TABLE 1. FOCUS MARKERS IN SAISIYAT (YEH 2000)
focus markers group i group ii
agent focus (af) m-, -om-, ma-, Ø Ø
patient focus (pf) -ºn -i
locative focus (lf) -an
referential focus (rf) si- -ani
Choice of negator is subject to cooccurrence restrictions for focus markers in Groups I and II, the details of which will be explained in section 2.3. Examples (1) through (4)8 illustrate the use of different types of focus markers in af²rmative sentences.
example of af (data from our ²eld work) (1) Ø ¿oja¿ s-om-ºt ka korkori¥.
nom mother beat-af-beat acc child
‘Mother beats the child.’
example of pf (data from our ²eld work) (2) Korkori¥ ni¿ ¿oja¿ sººt-ºn.
child gen mother beat-pf
‘The child is beaten by mother.’
example of lf (data from Yeh 2000) (3) Ka-patol-an hini¿.9
nmlz-sing-lf this
‘This is the place for singing.’
example of rf (data from Yeh 2000) (4) Kahœj si-sººt ni aki¿ ka korkori¥.
stick rf-beat gen grandfather acc child
‘The stick was used by grandfather to beat the child.’
In (1), adding the af in²x -om- to the verb sººt10‘beat’ places focus on the agent ¿oja¿ ‘mother,’ marking “mother” as the agent in the sentence ‘Mother beats the child.’ When a verb takes the pf suf²x –ºn, as in (2), the patient becomes the focus of the sentence, which marks “child” as the patient subject in ‘The child is beaten by mother.’ In (3), the lf suffix -an and nominalization marker ka- are both added to the verb patol ‘sing’, creating the word kapatolan ‘the place for singing’; this construction places sentential focus on the location of singing. In (4), the use of the rf (referential) suf²x places the focus on the instrument kahœj ‘stick’.
Syntactic position can also be used to focus salient information (see Yaeger-Dror 2002, Cutler et al. 1997, Fowler and Housum 1987, Kadmon 2001, Koopmans-van Beinum 1992). In Saisiyat, utterance-initial position, which might be a prosodically focused syntac-tic position, often cooccurs with a morphosyntacsyntac-tically marked focused NP. Because Saisiyat typically exhibits SVO word order,11 the focused argument (as a sentential sub-ject) occurs at or near the beginning of a sentence. Thus, it is possible that a syntactically or positionally de²ned item such as a focused agent12 may realize prosodic prominence.
The results of this study indicate that sentential subjects are the most acoustically prominent items in the Saisiyat negative experimental sentences. This contrasts sharply with our English experimental sentences, in which the negators themselves are the most acoustically prominent items. In addition, the presence of negators does not
8. The phonemic inventory of Saisiyat is introduced in 2.1.
9. Yeh (2000) pointed out that this construction is usually used in equational sentences. Please note that hini¿ ‘this’ in Yeh (2000) is pronounced as hini by our informant.
10. Based on our data drawn from ²eldwork, we ²nd that sººt ‘beat’ can have ººt as a variant.
11. Verb-initial word order has also often been observed in natural discourse of Saisiyat. 12. This study’s test materials include only agent-focused subjects.
signi²cantly change the prosodic parameters of contiguous words in Saisiyat. English, in contrast, exhibits relatively large-scale prosodic differences between af²rmative and negative sentences. We propose that the following factors can account for the cross-lin-guistic variation observed in our data: (1) the relationship between prosodic promi-nence and syntactic subjects in Saisiyat, (2) transparency ofthe Saisiyat negation system, and (3) the relationship between prosodic prominence and semantically de²ned focus in English.
The remainder of this paper is organized in the following way: First, we provide a brief introduction to the phonological, morphological, and syntactic structures of Saisiyat. Then, we describe the method and results of Experiments 1 and 2, in which we measured the relative prosodic prominence of lexical items in both af²rmative and negative sentences in both Saisiyat and English. Then, to investigate the question of whether the presence of a negator would result in global prosodic changes in either lan-guage, we compare our acoustic measurement of the items surrounding negators in af²rmative and negative sentences between Saisiyat and English. Finally, we provide possible interpretations for our results and suggest directions for future research.
2. THE STRUCTURE OF SAISIYAT. This section provides a brief introduction
to the phonology, morphology, and syntax of Saisiyat. This information has three sources: Yeh (2000), our own ²eldwork, and a database of Saisiyat discourse collected at the Graduate Institute of Linguistics of National Taiwan University.
2.1 THE PHONEMIC INVENTORY OF SAISIYAT. The phonemic
inven-tory of Saisiyat consists of seventeen consonants13 and six vowels; they appear in tables 2 and 3.14
The accent within words usually falls on the last syllable, except for function words and place names.15 The most prevalent syllable structures are CV, CVC, and CVV. According to Chiang and Chiang (2005), Saisiyat can be classi²ed as a pitch accent language at the lexical level, because lexical accent is realized by means of speci²c F0 patterns, rather than by duration and intensity.
2.2 SAISIYAT CASE MARKING. In contrast with other Austronesian
lan-guages of Taiwan, which exhibit VOS word order, Saisiyat exhibits primarily SVO word
13. Note that /s/ and /z/ are phonemes of Northern Saisiyat that correspond to /‰/ and /ð/ in South-ern Saisiyat. For consistency, these two variants are both represented as /s/ and /z/ throughout the paper. Moreover, the lateral ³ap was reported to be extinct in some dialects and retained by only a few speakers (Yeh 2000). For this reason, this sound is not included in table 2. With respect to the Saisiyat vowel inventory, the presence of /u/ is controversial, so we do not include it in our inventory.
14. The letters in parentheses represent the ASCII equivalents of IPA characters, following the transcription system that appears in Yeh (2000). For consistency, all Saisiyat words will be transcribed in standard IPA.
15. Yeh (2000) indicated that the ultimate stress rule does not apply to function words or place names. We observed that the prosody of function words in Saisiyat was much less regular than that of content words. Many function words are either monosyllabic or disyllabic and bear neither stress nor accent, unless they are emphasized for pragmatic reasons. As for place names, too few tokens occurred in our data to discern any patterns.
order. Case markers usually occur before nouns to mark their syntactic function. Accord-ing to Yeh (2000), there are six case markers in Saisiyat, each of which is divided into two categories: persons (not including pronouns) and common nouns, as shown in table 4. In example (5), the subject jaa¿ ‘father’ is assigned nominative case, which is marked with the zero morpheme, and ka marks korkori¥ ‘child’ as accusative case.
(5) Ø Jaa¿ -om-ºt ka korkori¥.
nom father beat-af-beat acc child
‘Father beats the child.’ (data from Yeh 2000)
2.3 NEGATORS IN SAISIYAT. Yeh et al. (1998) and Yeh (2000) claim that
Saisiyat has an inventory of eight negators: ¿okik, ¿okaj, ¿amkik, ¿amkaj, kajni¿, ¿oka¿, ¿izi¿, and ¿in¿ini¿.16Negators are chosen from this inventory according to the syntactic constructions in which they appear, and they are followed by verbs with focus markers chosen according to sentential focus. Examples (6) to (13) describe the distribution and cooccurrence restrictions of the eight negators in Saisiyat, based on the descriptions given in Yeh (2000). All data are from Yeh (2000).
16. See Zeitoun (2001) for a different analysis of some negators.
TABLE 2. CONSONANTS IN SAISIYAT
bilabial alveolar alveopalatal palatal velar glottal
stop p t k ¿ (’) nasal m n ¥ (ng) fricative vl. s S (S) h fricative vd. B(b) z lateral l trill r glide w j (y)
TABLE 3. VOWELS IN SAISIYAT
front central back
high i
mid œ (oe) º (e) o
low Q (ae) a
TABLE 4. THE CASE SYSTEM OF SAISIYAT (YEH 2000)
nominative accusative genitive possessive dative locative personal name Ø
hi hi ni /an-a /ini kankala
common nouns Ø
(6) ¿okaj is followed by dynamic verbs, which must be af²xed with Group II focus markers; these cannot occur with aspectual markers.
¿oja¿ ¿okaj ººt ka korkori¥.
mother neg beat-Ø acc child
‘Mother does not beat the child.’
(7) ¿okik is followed by nominal predicates, stative verbs or dynamic verbs af²xed with aspectual markers.
Sia ¿okik sarara¿ jakin.
he neg like-Ø me
‘He dislikes me.’
(8) ¿amkaj cooccurs with future events or possible events. It is a blended form derived from ‘¿am’ (a modal verb similar to the English ‘will’) and ‘¿okaj.’ Like ¿okaj, it is followed by dynamic verbs with Group II focus markers, and it cannot cooccur with aspectual markers.
aki¿ ¿amkaj wa:i¿.
grandfather neg come-Ø
‘Grandfather will not come.’
(9) ¿amkik cooccurs with future events or possible events. It is a blended form derived from ‘¿am’ and ‘¿okik.’ Like ¿okik, it is followed by nominal predi-cates, stative verbs, or dynamic verbs af²xed with aspectual markers. Sia rim¿an ¿amkik raj taw¿an.
he tomorrow neg loc home
‘He will not be at home tomorrow.’
(10) kajni¿ is used in volitional constructions. It can be followed by nouns or verbs with Group I focus markers.
Jako kajni¿ m-a¿rºm.
I neg af-sleep
‘I don’t want to sleep.’
(11) ¿oka¿ occurs in existential constructions; it can only be followed by nouns. Jako ¿oka¿ ka rajhil.
I neg acc money
‘I have no money.’
(12) ¿izi¿ is an imperative negator, similar to English ‘don’t.’ The verb following it is af²xed with Group II focus markers.
¿izi¿ ººt ka korkori¥.
neg beat-Ø acc child
‘Don’t beat the child.’
(13) ¿in¿ini¿ ‘not yet’ precedes adverbs or verbs with Group II focus markers. Jako ¿in¿ini¿ o:næhne: m-wa:i¿.
I neg for-a-long-time af-come
The above descriptions and examples show that negators in Saisiyat have many dif-ferent forms and functions, and that they must cooccur with speci²c syntactic or semantic structures. Sentence (14) contains an example of an additional negator found in our data, which appears to be a modal with a meaning similar to the English ‘cannot.’
(14) ¿œaj haa¿ m-atol.
’Œbaj cannot af-sing
‘’Œbaj cannot sing.’
3. EXPERIMENTS FOR TESTING YAEGER-DROR’S “COGNITIVE PROMINENCE PRINCIPLE”
3.1 EXPERIMENT 1. Experiment 1 compared the prosodic prominence of items
in both Saisiyat and English sentences, including sentential subjects, negators and words directly following negators (hereafter referred to as ‘X’). For the purpose of comparison with negative sentences, the prosodic prominence of subjects and X-constituents was also compared in af²rmative sentences.
3.1.1 Methodology17
3.1.1.1 Informants. Three informants participated in the experiment: two native
Saisiyat speakers and one native UK-English speaker. One of the Saisiyat informants speaks Northern Saisiyat and the other speaks Southern Saisiyat.18 Both are male, and between 50 and 60 years of age. They also speak Japanese and Hakka, a Chinese dia-lect spoken in Taiwan. The British English–speaking informant is a 24-year-old male from London. By self-report, none of the informants had a problem related to either hearing or articulation. Recordings were made in the speech lab at the Graduate Insti-tute of Linguistics at National Taiwan University using a Kay Elemetrics CSL 4400. A condenser microphone was placed approximately 10 centimeters away from the infor-mants’ mouths for the duration of the recording. Total recording time for each partici-pant was approximately one hour.
3.1.1.2 Materials and procedure. Saisiyat is a language without a writing
sys-tem, so it was not possible to elicit the sentences using written materials. Instead, infor-mants were asked questions by the researcher, and they were instructed to answer each of the questions in the af²rmative or in the negative. Examples are given in (15a–b). The experimental materials included 15 pairs of elicited af²rmative and negative sentences in both English and Saisiyat.
17. Aside from the content of the experimental materials, the methodology given for Experiment 1 applies to Experiment 2. Therefore information on informants, procedure, and measurement will be given for Experiment 1 only and will not be repeated in the description of Experiment 2. 18. Because Saisiyat is an endangered language, informants who are both pro²cient enough to
participate in the experiment and ³uent enough in Mandarin Chinese to comprehend the instructions were very dif²cult to ²nd.
elicitation of affirmative sentences (15) a. Researcher: ¿œaj minatini¿ aj?19
’Œbaj brother particle
‘Is ’Œbaj (someone’s) brother?’ Informant: ¿œaj minatini¿.
’Œbaj brother
‘’Œbaj is (someone’s) brother.’
ELICITATIONOFNEGATIVESENTENCES
b. Researcher: ¿œaj minatini¿ aj?
’Œbaj brother particle
‘Is ’Œbaj (someone’s) brother?’ Informant: ¿œaj ¿okik minatini¿.
’Œbaj neg brother
‘’Œbaj is not (someone’s) brother.’
English sentences were elicited using the same procedure, as exempli²ed in (16a–b):
ELICITATIONOFAFFIRMATIVESENTENCES
(16) a. Researcher: Is Bob your brother? Informant: Bob is my brother.
ELICITATIONOFNEGATIVESENTENCES
b. Researcher: Is Bob your brother? Informant: Bob is not my brother.
Each Saisiyat negative sentence contained a single subject, a negator, and a predicate. The predicate consisted of either a noun or a stative/dynamic verb, depending on the negator chosen. The grammatical subject of each sentence was ¿œaj, a common name for men. Saisiyat experimental materials included the following bisyllabic negators: ¿okik, haa¿, ¿okaj, ¿oka¿, and kajni¿.20 Words appearing after negators were controlled for segmental content; these were limited to words composed exclusively of sonorants and vowels, so that an uninterrupted F0 track could be extracted from that area. Our Northern Saisiyat informant con²rmed that all experimental sentences were grammat-ical and acceptable.
To avoid the possible confound of semantic discrepancy between the two sets of materials, the English experimental sentences were designed to be translations of the Saisiyat sentences. In the English materials, however, the subject’s name was changed from “’Œbaj” to “Bob,” as given in (17).
ENGLISHSENTENCES
(17) a. Bob is my brother. (af²rmative) b. Bob is not my brother. (negative)
19. In (15a–b), the presence of the interrogative particle aj marks these sentences as interrogative. 20. We chose bisyllabic negators to better match bisyllabic subjects in experimental materials. Among bisyllabic negators, we did not include ¿amkik and ¿amkay, which have been classi²ed as negators blended with the future tense marker ¿am (Yeh 2000). We did not include ¿izi¿, because it is an imperative negator, which would have disrupted SVO word order. Also excluded was the trisyl-labic negator ¿in¿ini¿ ‘not yet,’ because its relatively complex syntactic structure would have made the sentence in which it appeared differ substantially from the other experimental items.
Note that negation in both the Saisiyat and English materials was designed not to elicit a narrow-focus contrastive reading (e.g., ’Œbaj is not X’s brother, but he could be someone else’s), but to have a broad-focused reading over the predicate. The subjects were instructed to answer the negative sentences with a broad-focused reading in both languages. After the informants had answered one question, both the researcher and the informant listened to the recorded response to determine whether it had been ³uently and naturally pronounced. If not, or if the informant had hesitated, he was asked to read it again, until the utterance had been produced to the informant’s satisfac-tion. Then, the informant was instructed to proceed to the next item.
3.1.1.3 Measurement of prosodic prominence. Three prosodic parameters are
generally used to realize prominence on a focused or an accented word: fundamental frequency (F0), higher intensity, and increased duration (see Fox 2000 for a detailed discussion). For our data, the F0 peak, intensity peak, and duration of thesyllable rime (the vowel and coda consonant) appearing in accented syllables were measured using Praat 4.1.19 signal processing software. In the Saisiyat materials, accented syllables of content words were invariably ²nal syllables. For example, in the sentence ¿œaj ¿okik minatini¿ ‘’Œbaj is not (someone’s) brother,’ we measured the F0 peak and intensity peak in the rime of the last syllable of each word (namely -aj, -ik, -i¿), because accent in Saisiyat falls on the ²nal syllable. Figure 1 shows the measured area for the word
FIGURE 1. AN EXAMPLE OF F0 PEAK MEASUREMENT FOR THE SAISIYAT WORD ’ŒBAJ (A PROPER NAME)†
¿œaj, with the arrow indicating peak location. The arrow in figure 2 marks the inten-sity peak for the measured portion of the word ¿œaj.
Syllable onsets were excluded from these measurements to avoid the possible con-found of microprosodic variation that would have been introduced by intrinsic F0 dif-ferences in initial consonants. Furthermore, the accusative marker ka, which occurs in existential sentences, was also excluded from measurement, because it is unaccented when it appears in isolation. Instead, the accented word following it was measured.
In English, we measured the same three parameters in syllable rimes for the follow-ing items in negative sentences: the subject “Bob,” the negator “not,” and the ²rst sylla-ble of the word (X) that followed the negator. We also measured the same parameters for the subject “Bob” and the X-constituent in af²rmative sentences, for the purpose of comparison across utterance types.
3.1.2 Results. Table 5 summarizes comparison among the Northern Saisiyat,
South-ern Saisiyat, and English informants with respect to their production of both F0 and inten-sity peaks. It shows general patterns in af²rmative and negative sentences for the Northern Saisiyat informant. In af²rmative sentences, the subject ¿œaj is realized with a signi²cantly higher F0 peak and intensity peak than X is (t[14]=6.708, p<.01 for F0 peak, and t[14]=5.123, p<.01 for intensity peak). In negative sentences, F0 peak and intensity peak are also realized on the sentential subject. The mean difference between item
cate-FIGURE 2. AN EXAMPLE OF INTENSITY PEAK MEASUREMENT FOR THE SAISIYAT WORD ’ŒBAJ†
gories was found to be signi²cant in a one-way anova (F[2,42]=7.27, p<.01 for F0 peak, and F[2,42]=9.986, p<.01 for intensity peak). A posthoc test indicated that the difference between ¿œaj and X contributes to the contrast in F0 peak mean (t[42]=2.17, p<.01), while the contrast in intensity peak mean is created by ¿œaj being signi²cantly higher than both negator and X (t[42]=3.72, p<.01, t[42]=4.002, p< .01). 21
TABLE 5. COMPARISON OF THE F0 AND INTENSITY PEAKS
IN SUBJECT, NEGATOR, AND X-CONSTITUENTS BETWEEN AFFIRMATIVE AND NEGATIVE SENTENCES affirmative sentences subject x t-test
Northern
Saisiyat F0 peak (Hz) meansd 133.227.08 121.68 t(14)=6.708, 6.88 p<.01 intensity peak (dB)mean 78.55 75.08 t(14)=5.123, p<.01
sd 1.35 2.91
Southern
Saisiyat F0 peak (Hz) meansd 115.645.41 99.61 t(14)=6.480, 9.93 p<.01 intensity peak (dB)mean 76.75 71.97 t(14)=9.957, p<.01
sd 3.10 2.67 English F0 peak (Hz) mean 132.76 134.42 t(14)=-.317, p=.756 sd 8.25 14.82 intensity peak (dB) mean 77.61 70.75 t(14)=7.061, p<.01 sd 1.83 3.91
negative sentences subject negator x anova
Northern
Saisiyat F0 peak (Hz) meansd 134.898.17 127.112.29 121.29 F(2, 42)=7.27,8.39 p<.01 intensity peak (dB)mean 78.43 75.35 75.12 F(2, 42)=9.986, p<.01
sd 2.69 2.34 1.64
Southern
Saisiyat F0 peak (Hz) meansd 115.737.08 107.217.23 100.43 F(2, 42)=12.772,10.23 p<.01 intensity peak (dB)mean 76.4 72.35 70.83 F(2, 42)=7.551, p<.01
sd 3.61 4.44 4.09 English F0 peak (Hz) mean 130.77 142.73 110.69 F(2, 42)=122.816, p<.01 sd 4.67 5.17 6.89 intensity peak (dB) mean 77.24 72.63 68.5 F(2,42)=46.147, p<.01 sd 2.24 1.63 3.31
21. Duration data have been temporarily excluded from this section, because our Saisiyat and English sentences contain words that have different nuclear vowels, syllable structures, and syllabicities. The variations in duration caused by these factors would have been likely to dis-tort our overall analysis. Yaeger-Dror (2002:1497) found that “many factors in³uence dura-tion” and “while duration was coded, it has not been a useful measure.” Duration measurements have been included in section 4, where tokens of the same item were compared between af²rmative and negative sentences within the same language. The results of Experi-ment 2 (see 3.2) also include Saisiyat duration data for Saisiyat experiExperi-mental itemswhose rimes were designed to be uniform in duration.
The Southern Saisiyat informant’s data display patterns similar to those of the Northern Saisiyat informant. Both F0 and intensity peaks are realized on the sentential subject in af²rmative sentences (for F0 peak: t[14]=6.480, p<.01; for intensity peak: t[14]=9.957, p<.01). Similarly, the three items in negative sentences are signi²cantly different with respect to both parameters (for F0 peak: F[2,42]=12.772, p<.01; for intensity peak: F[2,42]=7.551, p<.01). A posthoc test revealed that the signi²cant dif-ference emerges as the result of difdif-ferences between the subject and X (F0 peak: t(42)=5.04, p<.01; intensity peak: t(42)=3.758, p<.01).
The English informant’s data with respect to intensity peak are similar to the Saisiyat informants’ data; the highest intensity peak is located in the sentential subject for both af²rmative and negative sentences (af²rmative: t[14]=7.061, p<.01; negative: F[2,42]=46.147, p<.01). However, the English informant’s data differ with respect to F0 peak. In af²rmative sentences, the sentential subject and X have F0 peaks at similar values, which show no signi²cant difference. In negative sentences, the highest F0 peaks are realized on negators (142.73 Hz) and the values among three items are signi²cantly different (F[2,42]=122.816, p<.01). A posthoc test showed that the mean differences in F0 and intensity peak among the three item types are all signi²cant (F0 peak: subject vs. negator: t[42]=5.79, p<.01; subject vs. X: t[42]=9.72, p<.01; negator vs. X: t[42]=15.51, p<.01; intensity peak: subject vs. negator: t[42]=5.06, p<.01; sub-ject vs. X: t[42]=9.60, p<.01; negator vs. X: t[42]=4.54, p<.01).
To investigate the possibility of individual variation among sentences, we calculated the occurrence frequency of highest F0 peak and intensity peak on each item in a sentence separately for all experimental sentences; these calculations appear in table 6. As dis-played in the table, Saisiyat and English differ from each other with respect to F0 peak realization in negative and af²rmative sentences. For the two Saisiyat speakers, the high-est F0 and intensity peaks in sentences are mostly realized on sentential subjects (F0 peak:
TABLE 6. F0 AND INTENSITY PEAK LOCATION FOR THREE ITEM TYPES
IN AFFIRMATIVE AND NEGATIVE SENTENCES affirmative sentences
speaker prosodic parameters subject x
Northern Saisiyat highest F0 peak (Hz) 100% 0%
highest intensity peak (dB) 100% 0%
Southern Saisiyat highest F0 peak (Hz) 100% 0%
highest intensity peak (dB) 100% 0%
English highest F0 peak (Hz) 78.6% 21.4%
highest intensity peak (dB) 100% 0%
negative sentences
speaker prosodic parameters subject negator x
Northern Saisiyat highest F0 peak (Hz) 80% 20% 0%
highest intensity peak (dB) 100% 0% 0%
Southern Saisiyat highest F0 peak (Hz) 100% 0% 0%
highest intensity peak (dB) 93.3% 6.7% 0%
English highest F0 peak (Hz) 0% 100% 0%
from 80–100 percent; intensity peak: from 93.3–100 percent). Exceptions are found in the Northern Saisiyat speaker’s negative sentences 4, 5, and 6, in which the negator haa¿ received the highest F0 peak. An exception in intensity peak location occurs in the South-ern Saisiyat speaker’s negative sentence 12, with his highest intensity peak falling on the negator ¿oka¿.22
The English speaker, in contrast, realizes 78.6 percent of the highest F0 peaks on sentential subjects in af²rmative sentences, and 100 percent of his highest F0 peaks on the negator in negative sentences. Intensity peaks are realized on sentential subjects in all experimental sentences (100 percent).
3.2 EXPERIMENT 2: -aj SENTENCES. The English data analyzed in
Experiment 1 provide evidence to support the Cognitive Prominence Principle, because the English informant invariably realized F0 peaks on negators.
In Experiment 2, to investigate the possibility that microprosodic effects of the -aj rime in the subject ¿œaj skewed the results of Experiment 1, another 15 sentences were constructed for Saisiyat, each word of which has an -aj rime, an example of which appears in (18).23
(18) Researcher: ¿œaj º:aj aj?
’Œbaj give particle
‘Does ’Œbaj give (something)?’ Informant: ¿œaj ¿okaj º:aj.
’Œbaj neg give
‘’Œbaj does not give (something).’
If we obtain similar results holding vowel quality constant across item types, it diminishes the likelihood that aj is simply intrinsically higher in F0 and/or intensity than other vowels. Furthermore, because each of the tested items in Experiment 2 contains the same rime, duration comparisons can be made across items.
3.2.1 Results. As can be seen in table 7, the subject ¿œaj remains the most
promi-nent item for all parameters measured except duration, even when the last threeitems were controlled for vowel quality. In Experiment 2, F0 peak and intensity peak location measurement results are similar to those of Experiment 1; peaks fall mostly on senten-tial subjects (86.6–100 percent) in both af²rmative and negative sentences. F0 peak is invariably realized on the subject for both Saisiyat speakers. Intensity peaks are also most often realized on sentential subjects, with three exceptions: the negator in the ninth sentence and the X-aj in the ²fteenth sentence for the Northern Saisiyat speaker, and the X-aj in the sixth sentence for the Southern Saisiyat speaker.24
A posthoc test showed that the signi²cance of both F0 and intensity peak results stems from the mean difference between ¿œaj and ¿okaj, ¿œaj and X-aj. That is to say, the F0 and intensity peaks of ¿œaj are signi²cantly higher than ¿okaj (F0 peak: t[42]=4.91, p<.01; intensity peak: t[42]=4.03, p<.01) and X-aj (F0 peak: t[42]=6.94, p<.01; intensity peak: t[42]=4.48, p<.01).
22. Because the percentage of exceptions is low, we can ²nd no systematic explanation to account for them. We regard the F0 prominence of haa¿ for Northern Saisiyat speakers as an idiosyncratic variation. 23. A complete list of sentences is given in appendix B.
The Southern Saisiyat informant’s data are similar to those of the Northern Saisiyat informant. The subject ¿œaj is still the most prominent in terms of both F0 and inten-sity peaks. In addition, a posthoc test revealed that the subject ¿œaj is signi²cantly higher than both ¿okaj (F0 peak: t[42]=6.07, p<.01; intensity peak: t[42]=5.57, p<.01) and X-aj (F0 peak: t[42]=5.68, p<.01; intensity peak: t[42]=3.45, p<.01).
In terms of rime duration, however, the rimes of sentential subjects are not found to be longer than those of other constituents. The mean rime duration for sentential subjects is in between that of the negator and the X-aj constituent for the two Saisiyat speakers. In the Northern Saisiyat informant’s case, the negator has the longest rime duration (0.403 seconds). A posthoc test demonstrated that the overall difference (F[2, 42]=10.255, p<.01) is the result of the signi²cant contrast between the negator and the X-aj constitu-ent. In the Southern Saisiyat informant’s case, X-aj has the longest duration (0.274 sec-onds); in this case, it is the contrast between X-aj and the negator that produces the overall difference (F[2, 42]=5.993, p<.01).
4. COMPARISON OF AFFIRMATIVE AND NEGATIVE SENTENCES IN SAISIYAT AND ENGLISH. In this section, we investigate the prosodic differences
between af²rmative and negative sentences, comparing the F0 peak, intensity peak, and duration of sentential subjects and X constituents. In addition, we compare af²rmative and negative sentences in terms of subject-X pair differences, to determine whether the
TABLE 7. LOCATION OF F0 AND INTENSITY PEAK
FOR SUBJECT, NEGATOR, AND X IN NEGATIVE SENTENCES (THREE ITEM RIMES CONTROLLED AS –aj)
northern saisiyat subject negator x anova
pitch peak (Hz) mean 134.99 118.16 111.2 F(2, 42)=25.496,
p<.01
sd 9.57 11.95 5.45
highest F0†
† Frequencies of cases that carry the highest F0 in the negative sentence.
100% 0% 0%
intensity peak (dB) mean 65.94 63.34 63.05 F(2, 42)=12.185,
p<.01
sd 1.27 1.39 2.41
highest intensity ‡
‡ Frequencies of cases that carry the highest intensity in the negative sentence.
86.6% 6.7% 6.7%
duration (sec) mean 0.364 0.403 0.335 F(2, 42)=10.255,
p<.01
sd 0.042 0.038 0.042
southern saisiyat subject negator x anova
pitch peak (Hz) mean 136.93 121.49 122.49 F(2, 42)=23.084,
p<.01
sd 4.71 7.2 8.45
highest F0† 100% 0% 0%
intensity peak (dB) mean 71.69 67.95 69.37 F(2, 42)=15.812,
p<.01
sd 1.51 1.46 2.40
highest intensity ‡ 93.3% 0% 6.7%
duration (sec) mean 0.267 0.226 0.274 F(2, 42)=5.993,
p<.01
appearance of a negator in a sentence will affect the F0 drop-off (declination) of its intona-tional contour. Second, we compare the difference between the F0 height changes in sub-ject-negator pairs with those in negator-X pairs for both English and Saisiyat negative sentences. If differences exist, it is possible that the presence of a negator causes different kinds of global intonational modi²cation in different languages.
4.1 METHODOLOGY. The original 15 pairs of Saisiyat and English negative
and af²rmative sentences used in Experiment 1 were reanalyzed to compare the F0 peak, intensity peak, and syllable rime duration of sentential subjects and Xs between af²rmative and negative sentences. Example sentences appear below in (19) and (20).
SAISIYATSENTENCES
(19) a. Af²rmative: ¿œaj minatini¿.
’Œbaj brother
‘’Œbaj is (someone’s) brother.’ b. Negative: ¿œaj ¿okik minatini¿.
’Œbaj neg brother
‘’Œbaj is not (someone’s) brother.’
ENGLISHSENTENCES
(20) a. Af²rmative: Bob is my brother. b. Negative: Bob is not my brother.
In addition, pair differences between the same item appearing in an af²rmative and a negative sentence were calculated as:
a. F0 peak difference: | F0 peakA – F0 peakB | (in Hz)
b. intensity peak difference: | intensity peakA – intensity peakB | (in dB) The following pair differences were calculated for both languages: (1) subject-X pairs in af²rmative sentences, (2) subject-negator pairs, and (3) negator-X pairs in negative sentences.
4.2 RESULTS
4.2.1 Sentential subjects in af²rmative and negative sentences. Analysis of
sentential subjects in the three informants’ af²rmative and negative sentences is given in table 8. The Northern Saisiyat informant’s data show that, although there are small differences between the prosodic parameter values of sentential subjects in af²rmative and negative sentences, none of those differences is statistically signi²cant. The results of the Southern Saisiyat informants’ data do not deviate from those of the Northern Saisiyat informant. Small differences in these prosodic parameters do exist, but none of them is statistically signi²cant. Furthermore, there is no statistically signi²cant dif-ference in the F0 and intensity of sentential subjects between af²rmative and negative sentences in English.
4.2.2 Xs in af²rmative and negative sentences. Results of the X item analysis
for the three informants’ data are given in table 9. Both the Northern and Southern Saisiyat informants’ data show no statistically signi²cant differences in X items
between utterance types. The English informant’s data, in contrast, exhibit statistically signi²cant differences in X items between utterance types for all parameters measured, with the exception of duration. F0 and intensity in X items are signi²cantly higher in English af²rmative sentences than they are in negative sentences.
TABLE 8. COMPARISON OF THE F0 PEAK, INTENSITY PEAK,
AND DURATION OF SENTENTIAL SUBJECTS BETWEEN AFFIRMATIVE AND NEGATIVE SENTENCES
northern saisiyat affirmative negative differencemean paired samplet-test mean sd mean sd F0 peak (Hz) 133.22 7.08 134.89 8.17 1.66 t(14)=–.852, p=.409 intensity peak (dB) 70.99 29.68 78.43 2.69 7.44 t(14)=–.954, p=.356 duration (sec) 0.33 0.05 0.37 0.04 0.45 t(14)=–2.837, p=.013
southern saisiyat affirmative negative differencemean paired samplet-test mean sd mean sd F0 peak (Hz) 115.64 5.41 115.73 7.08 0.935 t(14)=–.053, p=.958 intensity peak (dB) 76.75 3.10 76.40 3.61 0.35 t(14)=.524, p=.608 duration (sec) 0.27 0.04 0.28 0.04 0.007 t(14)=–0.559, p=.585
english affirmative negative differencemean paired samplet-test mean sd mean sd F0 peak (Hz) 132.76 8.25 130.77 4.67 1.989 t(14)=.733, p=.476 intensity peak (dB) 77.61 1.83 77.24 2.24 0.37 t(14)=.613, p=.550 duration (sec) 0.14 0.03 0.15 0.02 0.0015 t(14)=–0.191, p=.851
TABLE 9. COMPARISON OF F0 PEAK, INTENSITY PEAK, AND DURATION
OF ITEM X BETWEEN AFFIRMATIVE AND NEGATIVE SENTENCES
affirmative negative mean
difference paired sample t-test mean sd mean sd northern saisiyat F0 peak (Hz) 121.68 6.88 121.29 8.39 0.388 t(14)=0.203, p=.842 intensity peak (dB) 75.08 2.91 75.12 1.64 0.04 t(14)=0.953, p=.357 duration (sec) 0.29 0.12 0.29 0.1 0.0007 t(14)=–0.043, p=.966 southern saisiyat F0 peak (Hz) 99.61 9.93 100.43 10.23 0.818 t(14)=–0.313, p=.759 intensity peak (dB) 71.97 2.67 70.83 4.09 1.14 t(14)=1.477, p=.162 duration (sec) 0.24 0.08 0.24 0.08 0.002 t(14)=0.219, p=.830 english F0 peak (Hz) 134.42 14.82 110.69 6.89 23.729 t(14)=7.236, p<.01 intensity peak (dB) 70.75 3.91 68.50 3.31 2.25 t(14)=4.983, p<.01 duration (sec) 0.29 0.13 0.28 0.11 0.124 t(14)=0.449, p =.660
4.2.3 Global intonational change across utterance types. To investigate the
possibility that global intonational change is triggered by the presence of a negator, we provide a direct comparison of F0 and intensity values of subject and X between nega-tive and af²rmanega-tive sentences in table 10. A direct comparison of subject and X shows no signi²cant difference between their parameter values in af²rmative and negative sentences. This result is consistent with the comparisons presented in 4.2.1 and 4.2.2 for the two Saisiyat informants; their subject and X did not differ signi²cantly across sentence types. As for the English informant, the subject and X difference is relatively large (20.09 Hz in F0 peak and 8.74 dB in intensity peak), but this difference is not signi²cant across sentence types. The other two pair differences (subject-negator vs. negator-X) are compared within negative sentences; the results appear in table 11.
The results reveal distinctive differences in F0 peak between Saisiyat and English neg-ative sentences. For the Saisiyat informants, the level of F0 height change between sub-ject-negator and negator-X is quite similar (13.45 Hz and 13.07 Hz for Northern Saisiyat, 10.68 Hz and 9.20 Hz for Southern Saisiyat). For the English speaker, however, the degree of F0 height change differs signi²cantly between subject-negator and negator-X. The difference between negator and X (32.04 Hz) is larger than that between subject and negator (11.95 Hz) (t[14]=11.296, p<.01). As for intensity peak values, only the Northern Saisiyat informant shows a signi²cant difference between subject-negator and negator-X.
TABLE 10. SUBJECT-X PAIR DIFFERENCES IN F0 PEAK, AND INTENSITY
PEAK BETWEEN AFFIRMATIVE AND NEGATIVE SENTENCES
affirmative negative mean
difference paired sample t-test mean sd mean sd northern saisiyat F0 peak (Hz) 11.54 6.66 13.60 8.07 2.05 t(28)=–0.760, p=.453 intensity peak (dB) 3.67 2.33 3.94 1.81 0.27 t(28)=–0.356, p=.725 southern saisiyat F0 peak (Hz) 16.69 8.28 16.00 8.45 0.69 t(28)=0.226, p=.823 intensity peak (dB) 4.78 1.86 5.57 2.02 0.79 t(28)=–1.118, p=.273 english F0 peak (Hz) 12.94 15.26 20.09 6.89 7.14 t(28)=–1.652, p=.110 intensity peak (dB) 6.86 3.76 8.74 4.19 1.88 t(28)=–1.293, p=.206
TABLE 11. COMPARISON OF F0 PEAK AND INTENSITY PEAK
BETWEEN SUBJECT–NEGATOR AND NEGATOR–X PAIRS IN NEGATIVE SENTENCES
subject-negator negator-x mean
difference paired sample t-test mean sd mean sd northern saisiyat F0 peak (Hz) 13.45 8.42 13.07 7.88 0.38 t(14)= 0.117, p=.909 intensity peak (dB) 4.53 2.30 1.81 1.25 2.72 t(14)=4.433, p<.01 southern saisiyat F0 peak (Hz) 10.68 5.81 9.20 4.10 1.48 t(14)=0.778, p=.450 intensity peak (dB) 4.25 2.47 2.54 1.88 1.71 t(14)=2.069, p=.058 english F0 peak (Hz) 11.95 6.83 32.04 8.82 20.09 t(14)=11.296, p<.01 intensity peak (dB) 4.61 2.48 4.53 3.28 0.07 t(14)=0.071, p=.945
5. DISCUSSION AND CONCLUSION. The major ²ndings of this paper are
summarized in this section, and possible explanations are offered to account for the prosodic differences observed between Saisiyat and English.
5.1 CORRESPONDENCE BETWEEN PROSODIC PROMINENCE AND
SYNTACTIC SUBJECT IN SAISIYAT. Although negators have been shown to
be prosodically prominent with respect to F0 in our English data, they are not the most acoustically prominent items in Saisiyat negative sentences. Thus, our results show that the “Cognitive Prominence Principle” holds for English negators but not for those in Saisiyat. Saisiyat informants realize more acoustic prominence on sentential subjects in negative sentences than they do on any other items. Because the posthoc test in Experi-ment 1 showed that there are signi²cant differences among all three item types in English negative sentences, we suggest that there is a principled difference in prominence-ranking orders between English and Saisiyat negative sentences. The ranking order for English would be negator > subject > X, and for Saisiyat, subject > negator or X.
In addition, a disjunction between mean F0 peak and intensity peak can be observed in the English data. In Experiment 1, the highest mean F0 peak falls on English negators, but the highest mean intensity peak falls on sentential subjects. The English informant’s reli-ance on F0 to realize prominence on the negator con²rms previous ²ndings demonstrat-ing F0 to be the primary prosodic parameter used to express focal prominence in English (Bolinger 1958, and see Yaeger-Dror 2002:1497 for literature review).25
We propose, on the basis of our Saisiyat results, that syntactic position determines prosodic prominence in Saisiyat; prosodic prominence will tend to fall on the noun that appears as the sentential subject. Because SVO is the predominant word order in Saisiyat, the focused argument (the sentential subject) occurs at or near the beginning of a sentence. Although this study’s materials include only agent-focused sentences, Chiang’s (2005) data show that both the agent in agent-focused sentences and the patient in patient-focused sentences receive prosodic prominence.26 The high cooccur-rence of prosodic prominence and sentential subjects in Chiang (2005) provides fur-ther evidence for syntactically determined prominence in Saisiyat.27 It remains to be seen whether the Austronesian languages of Taiwan with VOS word order would exhibit patterns similar to those found in our Saisiyat data.
5.2 TRANSPARENCY OF THE NEGATION SYSTEM IN SAISIYAT.
It is possible that division of negation functions into several separate morphemes in Saisiyat may increase the transparency of the negation system. Thus, there is no need
25. Yaeger-Dror (2002:1497) claimed that speakers of Standard American English have different patterns in production of fundamental frequency and intensity: “Amplitude generally appears to covary with fundamental frequency.”
26. Chiang’s thesis (2005) was under the supervision of the ²rst author of this paper.
27. Because focused nouns usually occur early in Saisiyat sentences, one might argue that the “global utterance intonation contour” quite often exhibits declination effects, which could contribute to onsets having a higher pitch than offsets. Previous studies have demonstrated that the declination shown in declarative sentences does not necessarily make the ²rst stressed word more perceptually prominent than subsequent words, nor the last word less prominent than those preceding it. Because our study investigated production rather than perception, it remains to be seen whether acoustically prominent items are also perceptually prominent.
to increase articulatory effort by raising F0 to highlight negators. Recently, there has been a growing trend to incorporate functional considerations into phonological theory (e.g., Boersma 1998, Flemming 2004). For example, Flemming (2004) proposes that phonological contrast is achieved through competing interaction of the following three constraints: (1) maximizing the distinctiveness of contrasts, (2) minimizing articulatory effort, and (3) maximizing the number of contrasts in a language’s inventory. These three functional goals may come into con³ict. Applying Flemming’s constraints to the phenomena observed in Saisiyat, we propose that the number of contrasts in the Saisiyat negation system has already been maximized by its large inventory of negative morphemes, and that the distinctiveness of those contrasts has already been maximized by the cooccurrence restrictions on those morphemes, so no extra articulatory efforts need be made to prosodically highlight negators.
5.3 CORRESPONDENCE BETWEEN PROSODIC PROMINENCE
AND SEMANTICALLY DEFINED FOCUS IN ENGLISH. To investigate
whether the presence of a negator may trigger global intonational modi²cation, we made additional comparisons (described in detail in 4.2.3) whose three major ²ndings are summarized here: (1) The English negator becomes prominent in F0 and intensity through its contrast with the X item, while in Saisiyat, there is no signi²cant difference in the F0 values of negators and X items.28 (2) F0 drop-off from the sentential subject to X does not differ signi²cantly between af²rmative and negative sentences for Saisiyat informants. (3) There is a greater F0 height change from negator to X than from subject to negator in English, while in Saisiyat there is no signi²cant difference in the level of the two drop-offs. Figure 3 illustrates this point, providing the F0 mean for each item in Saisiyat and English af²rmative and negative sentences.
Figure 3 shows that the English negator does affect global intonational contours; its presence suppresses the F0 and intensity of the item(s) following it. Saisiyat, in contrast, demonstrates general declination throughout sentences, which is not interrupted by the presence of a negator. We propose that a semantically de²ned focus item, such as the English negator, requires more articulatory effort than a syntactically de²ned one does, because the former cannot rely on linguistic form to enhance its prominence. Thus,
28. Note that the intensity of the English negator is signi²cantly higher than that of X, although the subject is the most prominent item in the whole sentence.
FIGURE 3. F0 MEAN FOR EACH ITEM TYPE IN SAISIYAT AND ENGLISH
English speakers must not only raise F0 at the negator to increase its prominence, but also lower F0 at the item(s) following the negator in order to maximize acoustic contrast.
In sum, this paper offers typological explanations for the differences observed between Saisiyat and English. Saisiyat assigns prosodic prominence to focused senten-tial subjects, while English assigns prosodic prominence to semantically-de²ned focus items. Moreover, the morphosyntactic transparency of the Saisiyat negation system may allow for minimization of articulatory effort in production of negators. This study contributes to our overall knowledge and understanding of the range of possible strate-gies for realizing negation across languages; it also provides potential topics for future research, to test whether our proposal can be extended to describe the prosody of VOS-dominant Austronesian languages of Taiwan, or the prosody of other language types.
APPENDIX 1. SAISIYAT SENTENCES AND THEIR CORRESPONDING ENGLISH SENTENCES USED IN EXPERIMENT 1
saisiyat sentences english sentences 1 ¿œaj minatini¿. ‘Bob is my brother.’
¿œaj ¿okik minatini¿. ‘Bob is not my brother.’
2 ¿œaj lalaor. ‘Bob dozes.’
¿œaj ¿okik lalaor. ‘Bob does not doze.’
3 ¿œaj lijao¿. ‘Bob is rich.’
¿œaj ¿okik lijao¿. ‘Bob is not rich.’
4 ¿œaj m-atol. ‘Bob sings.’
¿œaj haa¿ m-atol. ‘Bob cannot sing.’
5 ¿œaj miltamako¿. ‘Bob smokes.’
¿œaj haa¿ miltamako¿. ‘Bob cannot smoke.’
6 ¿œaj lalœhaj. ‘Bob has fun.’
¿œaj haa¿ lalœhaj. ‘Bob does not have fun.’
7 ¿œaj majna:a¿. ‘Bob waits.’
¿œaj ¿okaj ajna:a¿. ‘Bob does not wait.’
8 ¿œaj mwa:i¿. ‘Bob comes to my place.’
¿œaj ¿okaj wa:i¿. ‘Bob has not come to my place.’
9 ¿œaj miltamako¿. ‘Bob smokes.’
¿œaj ¿okaj miltamako¿. ‘Bob does not smoke.’ 10 ¿œaj hajza¿ ka lapowar. ‘Bob has a guava.’
¿œaj ¿oka¿ ka lapowar. ‘Bob does not have a guava.’ 11 ¿œaj hajza¿ ka laro¿. ‘Bob has a persimmon.’
¿œaj ¿oka¿ ka laro¿. ‘Bob does not have a persimmon.’ 12 ¿œaj hajza¿ ka mona:. ‘Bob has a snail.’
¿œaj ¿oka¿ ka mona:. ‘Bob does not have a snail.’ 13 ¿œaj ¿am lapowar. ‘Bob wants a guava.’
¿œaj kajni¿ lapowar. ‘Bob does not want a guava.’ 14 ¿œaj ¿am laro¿. ‘Bob wants a persimmon.’
¿œaj kajni¿ laro¿. ‘Bob does not want a persimmon.’ 15 ¿œaj ¿am mona:. ‘Bob wants a snail.’
REFERENCES
Boersma, Paul. 1998. Functional phonology. The Hague: Holland Academic Graphics. Bolinger, Dwight L. 1958. A theory of pitch accent in English. Word 14:109–49. ———. 1978. Intonation across languages. In Universals of human language II, ed. by
Joseph Greenberg, 471–524. Stanford: Stanford University Press.
Chiang, Fang-mei. 2005. The study of prosody in Saisiyat. MA thesis, National Taiwan University.
Chiang, Wen-yu, and Fang-mei Chiang. 2005. Saisiyat as a pitch accent language: Evi-dence from acoustic study of words. Oceanic Linguistics 44:404–26.
Cutler, Anne, Delphine Dahan, and Wilma van Donselaar. 1997. Prosody in the com-prehension of spoken language. Language and Speech 40:141–201.
Flemming, Edward. 2004. Contrast and perceptual distinctiveness. In Phonetically-based phonology, ed. by B. Hayes, R. Kirchner, and D. Steriade, 232–76. Cam-bridge: Cambridge University Press.
Fowler, Carol A., and John Housum. 1987. Talkers’ signaling of ‘new’ and ‘old’ words in speech and listeners’ perception and use of the distinction. Journal of Memory and Language 26:489–504.
Fox, Anthony. 2000. Prosodic features and prosodic structure: The phonology of supra-segmentals. New York:Oxford University Press.
Hirschberg, Julia. 1990. Accent and discourse content: Assigning pitch accent in synthetic speech: The given/new distinction and deaccentability. In Proceedings of the Eighth National Conference on Arti²cial Intelligence II: 952–57. Cambridge, Mass.: MIT Press. ———. 1993. Pitch accent in context. Arti²cial Intelligence 63:305–40.
Horn, Larry. 2001. A natural history of negation. Chicago: University of Chicago Press. Huang, Shuanfan. 2002. The pragmatics of focus in Tsou and Seediq. Language and
Linguistics 3(4): 665–94.
Kadmon, Nirit. 2001. Formal pragmatics. Oxford: Blackwell.
Koopmans-van Beinum, J. Florien. 1992. The role of focus words in natural and in synthetic continuous speech: Acoustic aspects. Speech Communication 11:439–52.
APPENDIX 2. DESIGNED NEGATIVE SAISIYAT SENTENCES USED IN EXPERIMENT 2
no. saisiyat sentences english translations
1 ¿œaj ¿okaj º:aj. ‘’Œbaj does not give (something).’ 2 ¿œaj ¿okaj sanraj. ‘’Œbaj does not marry.’
3 ¿œaj ¿okaj ramramaj. ‘’Œbaj does not comfort (someone).’ 4 ¿œaj ¿okaj sowaj. ‘’Œbaj does not grip (something).’ 5 ¿œaj ¿okaj kipazaj. ‘’Œbaj does not cut the paddy.’ 6 ¿œaj ¿okaj kiowaj. ‘’Œbaj does not pick the fruit.’ 7 ¿œaj ¿okaj ¿Qlipowaj. ‘’Œbaj is not covered by a blanket.’ 8 ¿œaj ¿okaj inkonkonaj. ‘’Œbaj does not roll down.’ 9 ¿œaj ¿okaj rasiwazaj. ‘’Œbaj does not divorce.’
10 ¿œaj ¿okaj tisko-Qwhaj. ‘’Œbaj does not say anything wrong.’ 11 ¿œaj ¿okaj pasaj. ‘’Œbaj did not die.’
12 ¿œaj ¿okaj ¿as¿asaj. ‘’Œbaj does not waste (food or money).’ 13 ¿œaj ¿okaj ¿italaj. ‘’Œbaj has not stopped (walking).’ 14 ¿œaj ¿okaj alalaj. ‘’Œbaj was not excited.’ 15 ¿œaj ¿okaj kakowaj. ‘’Œbaj does not move.’
Morel, Mary-Annick. 1995. Valeur énonciative des variations de hauteur mélodique en français. French Language Studies 5:189–202.
O’Shaughnessy, D., and J. Allen. 1983. Linguistic modality effects on fundamental fre-quency in speech. Journal of the Acoustical Society of America 74:1155–71. Rau, Der-Hwa Victoria. 1992. A grammar of Atayal. Taipei: The Crane Publishing Co. Remijsen, Bert. 2003, April 11. Prosodic marking of negation. Message posted to http://
linguistlist.org/issues/14/14-1173.html.
Shibatani, Masayoshi. 1988. Passive and voice. Amsterdam: John Benjamins.
Wouk, Fay. 1996. Voice in Indonesian discourse and its implications for theories of the development of ergativity. Studies in Language 20:361–410.
Yaeger-Dror, Malcah. 1985. Intonational prominence on negatives in English. Lan-guage and Speech 28:197–230.
———. 2002. Register and prosodic variation: A cross language comparison. Journal of Pragmatics 34:1495–1536.
Yeh, Marie M. 2000. A reference grammar of Saisiyat [in Chinese]. Formosan Languages Series 2. Taipei: Yuan-liu Publishing Co.
Yeh, Marie M., Lilian M. Huang, Elizabeth Zeitoun, Anna H. Chang, and Joy J. Wu. 1998. A preliminary study on negative constructions in some Formosan lan-guages. In Selected Papers from the Second International Symposium on Languages in Taiwan, 79–110. Taipei: The Crane Publishing Co.
Zeitoun, Elizabeth. 2001. Negation in Saisiyat: Another perspective. Oceanic Linguistics 40(1): 125–34.
