The effects of prosodic boundaries on nasality in Taiwan Min
Ho-hsien Pan
Department of Foreign Languages and Literatures, National Chiao Tung University, 1001 Ta Hsueh Road, Hsinchu, 300, Taiwan
共Received 22 June 2006; revised 9 March 2007; accepted 12 March 2007兲
This study explores the effects of prosodic boundaries on nasality at intonational phrase, word, and syllable boundaries. The subjects were recorded saying phrases that contained a syllable-final nasal consonant followed by a syllable-initial stop. The timing, duration, and magnitude of the nasal airflows measured were used to determine the extent of nasality across boundaries. Nasal amplitudes were found to vary in a speaker-dependent manner among boundary types. However, the patterns of nasal contours and temporal aspects of the airflow parameters consistently varied with boundary type across all the speakers. In general, the duration of nasal airflow and nasal plateau were the longest at the intonational phrase boundary, followed by word boundary and then syllable boundary. In addition to the hierarchical influence of boundary strength, there were unique phonetic markings associated with individual boundaries. In particular, two nasal rises interrupted by nasal inhalation occurred only across an intonation phrase boundary. Also, unexpectedly, a word boundary was marked by the longest postboundary vowel, whereas a syllable boundary was marked with the shortest nasal duration. The results here support the hierarchical effect of boundary on both domain-edge strengthening and cross-boundary coarticulation.
© 2007 Acoustical Society of America. 关DOI: 10.1121/1.2722208兴
PACS number共s兲: 43.70.Aj, 43.70.Fq, 43.70.Bk 关BHS兴 Pages: 3755–3769
I. INTRODUCTION
Cross-linguistically, articulatory movements vary ac-cording to focus condition and boundary strength. Though focus and prominence are signaled by hyperarticulation, the effect of boundary can be implemented by either the weak-ening or strengthweak-ening of articulatory movements共Edwards, Beckman, and Fletcher, 1991; de Jong, 1995; Fougeron and Keating, 1996; Tabain, 2003b兲. These articulatory weakening and strengthening effects show gradient variations according to the strengths of adjacent boundaries.
Like traffic signs that control deceleration and accelera-tion of vehicular movements, boundary strength controls the duration and magnitude of articulatory movements in pre-and postboundary position. Though some languages prefer final strengthening in the preboundary position, such as phrase-final lengthening before an English IP boundary, other languages prefer initial strengthening in the postbound-ary position, such as phrase-initial lexical tonal strengthening in Taiwanese 共Pan, in press兲. Thus, it is more difficult to generalize the prosodic effect of boundary on surface articu-lation across languages. To fully explore the effect of bound-ary, diversified data need to be collected from languages with different rhythmic structures.
This study expands the scope of prosodic articulatory studies from intonation-based languages, such as English 共Cho, 2002, 2004, 2005, 2006兲 and French 共Fougeron and Keating, 1996; Fougeron, 2001; Tabain, 2003a b兲 to a tone-based language by investigating how nasal consonants in Taiwan Min共Taiwanese兲 are influenced by prosodic bound-aries.
A. Taiwanese nasal consonants and vowels
Min, a Chinese dialect, spoken in Fujian and Southeast Asia among Overseas Chinese who trace their roots back to Fujian, is spoken by 70% of the population in Taiwan. Tai-wan Min has around 16 oral vowels and 11 nasal vowels, depending on the dialect. Taiwanese phonotactic constraints require that nasal vowels occur after initial nasal consonants /m, n, G/, such as /m3/ “thing,” and that oral vowels occur after initial oral stops /b, g, l/, such as /bi/ “rice.” These phonotactic constraints have led researchers to devise vari-ous phonological rules stating the alternation between ho-morganic initial nasal and voiced stops. Based on auditory impressions with little empirical phonetic data, these rules change voiced stops into a homorganic nasal when followed by a nasal vowel, or vice versa, e.g., /b, l, +/ -⬎ 关m, n, G兴/_ 关nasal vowel兴 or /m, n, G / -⬎ 关b, l, +兴 /_ 关oral vowel兴 共Ting, 1985; Zhang, 1989; Cheng, 1968, 1973兲. It should be noted that there is no /d/ in Taiwanese and /l/ is considered to be a voiced stop due to its stoplike quality 共Zhang, 1989兲. Pre-liminary EPG data concerning /l/ collected by the author showed that speakers produced /l/ with an alveolar closure and with either a lateral or central release. Therefore, /l/ was included as an alveolar stop along with /b, +/.
Pan共2004兲 in an earlier airflow study found that Taiwan-ese initial voiced stops changed into homorganic nasals when preceded by a final nasal consonant. For example /b/ be-comes 关m兴 in /saG bi/ 关saG mi兴 “send rice.” This finding challenges the accepted phonological rules that govern the distribution of nasal vs oral vowels. Since oral vowels can occur after both initial nasals, as in /saG bi/ 关saG mi兴 “send rice,” and after initial voiced stop /be bi/关be bi兴 “purchase rice,” this contradicts the traditional description of a phono-tactic constraint in which oral vowels occur after initial
voiced stops and nasal vowels occur after initial nasals. Rel-evant to the current study is the fact that initial stops and vowels are not always nasalized to the same extent after a nasal consonant. The current investigation explores the ex-tent of nasalization from a nasal consonant across a bound-ary. If a boundary is strong, then less nasalization is expected to occur on the upcoming segment; if the boundary is weak, then more nasalization across the boundary is expected. As airflow is an effective means of capturing nasality patterns in Taiwanese, data relating to it will be used in this study.
B. Nasality and prosodic boundary
Studies on articulation and boundary encompass three major issues:共1兲 domain-edge strengthening; 共2兲 variation of articulation in relation to the strength of the adjacent bound-ary; and共3兲 variation of cross-boundary coarticulation in re-lation to the strength of the intervening boundary.
Turning to the first issue of domain-edge strengthening, previous studies have discovered hyperarticulation to be greater in domain-initial segments or domain-final segments than in domain-medial segments 共Fougeron and Keating, 1997; Krakow, 1999; Cho, 2002, 2004, 2006兲.
To capture domain-edge-strengthening effects on velic movement, Krakow共1989, 1993兲 in a velocontour study in-vestigated the coordination pattern between velic and lip movements, by placing a syllable initial nasal /m/ in either word-initial 关V#mV兴 or -medial 关V.mV兴 positions, and a syllable-final nasal /m/ in either word-final关Vm#兴 or medial 关Vm.V兴 positions. She measured the duration and amplitude of velic and lip raising/lowering, and found that, regardless of the position within a word, the offset of velic lowering was closely timed to the end of lip raising for initial nasals but to the onset of lip raising for final nasals. In other words, velic lowering began earlier for final /m/ than for initial /m/. The velic and lip coordination patterns are influenced by the boundary positions. In addition to coordination patterns, she also discovered that a syllable’s position within a word in-fluenced the magnitude and duration of the velic and lip movements. For instance, nasals in word-final positions were lower in velic displacements, higher in velic raising, and longer in duration for both lip lowering and raising than nasals in syllable-final 共and word-medial兲 positions. That is to say, syllable-final nasals are more hyperarticulated in word-final than in word-medial positions.
In Taiwanese, Hsu and Jun共1996a, 1996b兲 in a study on voice onset time共VOT兲 at the edge of Taiwanese tone groups discovered that VOT was longer in segments produced in the tone-group-initial position than in the tone-group-final posi-tion. So, one can surmise that evidence exists indicating that speakers of Taiwanese use articulatory strategies to indicate the existence of a boundary. However, little is known about the effect of boundary type on nasality in Taiwanese.
Turning to the second issue on boundary strengthening and articulation variations, it has been found that jaw, lin-gual, and lip movements vary from the most canonical to the least canonical form as boundaries vary from strong to weak in languages such as English, French, Tamil, and Korean 共Fougeron and Keating, 1997; Byrd et al., 2000; Fougeron,
2001; Cho and Keating, 2001; Tabain, 2003a, 2003b; Cho, 2005, 2006兲. Moreover, in French, velic lowering varies in a hierarchical manner at intonational phrase, accentual phrase, word, and syllable junctures. By calculating the amount of nasal airflow as an indicator of velic lowering, differences were found in a nasal airflow study on French initial /n/ 共Fougeron and Keating, 1996兲. In this study, one of two sub-jects differentiated three out of four levels of boundaries based on the extent of nasality during the production of the nasal consonant. However, the other male speaker failed to produce any difference between any of the boundaries 共Fougeron and Keating, 1996兲. Keating et al. 共2004兲 ex-tended this study by adding more speakers, and found that three out of four speakers produced differences in the amount of nasality in relation to boundary type in a speaker-dependent manner 共Keating et al., 2004兲. Both studies dis-covered that the stronger the preceding prosodic boundary, the lesser the amount of nasal airflow recorded from the ini-tial /n/. Following Fougeron and Keating’s study 共1996兲, which found nasal airflow to be an effective means to study the influence of boundary on postboundary nasality, the present study also uses nasal airflow to investigate the effect of boundary on nasality patterns in Taiwanese.
Turning to the third issue, the effect of boundary type on cross-boundary coarticulation, it has been found that antici-patory nasal coarticulation is affected by boundary type in English. MacClean 共1973兲 used frame-by-frame tracing of velic movements recorded on lateral high-speed cinefluoro-graphic film to observe the effects of syntactic boundaries on velic movements. McClean found that the presence of major syntactic boundaries 共including marked phrase, clause, or sentence boundaries兲 between two vowels in a CV#VN con-text delayed the onset of velic movements relative to the offset of the preceding initial vowels more than less-marked syntactic boundaries 共including word boundary兲. Since En-glish vowels are unspecified in terms of phonemic nasal fea-tures, the velum started to decline during the articulation of the preboundary vowel in CV#VN. The commencement of velic lowering in anticipation of the upcoming nasal conso-nant was found to be earlier in productions by English speak-ers after weak boundaries and later after strong boundaries 共MacClean, 1973兲.
Generally speaking, cross-boundary coarticulation weakens as the intervening prosodic boundary varies from weaker to stronger levels in the prosodic hierarchy. As articu-latory movement becomes more canonical next to a strong boundary, it is less likely to coarticulate with neighboring segments. For example, Cho共2004兲 reported greater V-to-V coarticulation resistance across stronger prosodic boundaries. So far there has been no study on the effect of boundary on cross-boundary carry-over nasalization. Hence, this study uses nasal airflow to investigate how cross-boundary carry-over nasalization from a final nasal to an initial voiced stop in Taiwanese varies according to the hierarchical strength of the intervening boundaries.
C. Speaker-dependent articulatory patterns
Before moving on to investigate the relationship be-tween boundary and nasality in Taiwanese, it should be noted
that, as previously mentioned in the airflow study on French nasality共Fougeron and Keating, 1996; Keating et al., 2004兲, speaker-dependent nasality patterns have already been widely observed. Furthermore, the effect of boundary type on jaw, lip, and acoustical parameters also varies from speaker to speaker共Tabain, 2003a, 2003b; Cho, 2005兲. For example, among the acoustical and articulatory data of five speakers producing English /#bi/ and /#ba/, it was found that all the speakers produced /a/ with the greatest jaw movement and /i/ with the least jaw movement at strong boundaries; however, the duration of lip opening during the articulation of a vowel varied among the same speakers. Only two speak-ers produced vowels with a longer lip opening after a strong boundary 共Cho, 2005兲. Tabain 共2003a兲 studied the acoustic patterns of French word-final /a#C/ sequences produced at an utterance, intonational phrase, accentual phrase, or word boundary and found that there was much variation among speakers. Speakers used different cues to indicate different prosodic strengths. For example, some speakers used vowel duration to distinguish four types of boundaries, i.e., utter-ance ⬎ intonational phrase ⬎ accentual phrase ⬎ word, whereas others classified the vowel duration before different boundaries into two groups, i.e., utterance ⫽ intonational phrase⫽ accentual phrase ⬎ word. Moreover, when looking at the spectral tilt for consonants, /#f, s, b/, all subjects grouped the boundaries at two levels; however, two subjects also made a distinction between word and higher level boundaries 共utterance ⫽ intonational phrase ⫽ accentual phrase⬎ word兲, whereas another subject made a distinction between utterance and lower level boundaries 共utterance ⬍ intonational phrase⫽ accentual phrase ⫽ word兲.
Keating et al. 共2004兲, in an EPG study on VOT and lingual contacts during the production of Taiwanese initial /t/ and /n/ after utterance, intonational phrase, accentual phrase, word, and syllable boundaries by two speakers, discovered that although there was a general trend for lingual peak con-tact to vary with boundary type, speaker-dependent patterns were nonetheless still observed. It was found that one speaker failed to vary the production of /t/ at all, whereas another speaker varied the production of /t/ in relation to the prosodic boundary strength.
Nearly all previous studies on articulation and boundary have discovered speaker-dependent articulatory patterns 共Fougeron and Keating, 1996; Keating et al., 2004兲. It is, therefore, likely that there are a certain number of patterns that a speaker can choose from, as well as some other factors which influence articulatory pattern in addition to boundary. The following is a discussion on the influence of focus, dec-lination, and speaking rate on velic movements.
D. Other factors influencing velic movements
Stress, declination, and speaking rate have been found to influence velic movements共Krakow, 1993; Bell-Berti et al., 1995兲. For example, stress affects velic height by enhancing the position of the velum. That is, the high velic position for an oral stop is higher, whereas the low velic position for a nasal stop is lower in stressed syllables共Vaissiere, 1988兲. In addition to stress, declination also affects velic movements.
It was discovered that velic displacement of nasals produced in words in serial position gradually decreased from the be-ginning to the end of a sentence. In other words, the earlier the nasal is in a sentence, the lower the position of the velum 共Bell-Berti and Krakow, 1991; Krakow, Bell-Berti, and Wang, 1995兲.
In addition to stress and declination, speaking rate also influences velic height. Bell-Berti and Krakow 共1991兲 in a study on velic height for /s/ and /n/ found that the faster the speaking rate is, the lesser the difference in velic height be-tween the oral and nasal consonants. Conversely, the slower the speaking rate, the larger the difference between the velic positions for /s/ and /n/. Kuehn 共1976兲 noticed that these effects are speaker dependent. Though both speakers in Kue-hn’s共1976兲 study reduced the distance of velic movement in fast speech, one speaker reduced the extent of velic lowering by raising the lowest velic position, whereas the other speaker reduced the distance of velic movement by lowering the highest velic position. Kent, Carney, and Severeid共1974兲 reported that as speaking rate increased from a conversa-tional to a rapid rate, speakers increased the velocity of velic movements, or reduced the extent of the velic movement, or used both strategies to produce the same sentence.
In addition to a control on the effects of focus and dec-lination, as previous studies have done, this study also adopts a control on global speaking rate in an attempt to minimize speaker-dependent variations, and to reveal the effect of boundary on nasality.
E. Objectives
After controlling the confounding factors, this study fol-lowed the line of study of the variations of velic movements within multiple prosodic domains. By using nasal airflow, a successful method used to record the effect of boundary on French initial nasals and on cross-boundary nasalization in Taiwanese, the present study explores the effects of different prosodic boundaries on 共1兲 domain-final nasality and 共2兲 cross-boundary nasalization.
II. METHOD A. Speakers
Three male native Taiwanese speakers participated in the experiments. They were students at National Chiao Tung University at the time of recording. The three speakers were also fluent in Mandarin and had received more than 10 years of ESL education.
B. Materials
Three types of contexts were used in the current study. The first was a baseline context which elicited utterances not ending with a nasal and, thus, no cross-boundary nasaliza-tion. The baseline data set consisted of syllable-final vowels followed by initial voiceless stops, abbreviated as Vptk 共vowel # /p/, /t/, /k/兲 or voiced stops, abbreviated as Vblg 共vowel # /b/, /l/, /g/兲. The syllable-final vowels and following stops were produced across intonational phrase, word, and syllable boundaries. As previously mentioned, /l/ was
in-cluded together with /b, +/ due to its stoplike quality.共2兲 The second context elicited comparison data with a final nasal followed by an initial voiceless stop that prohibits cross-boundary nasalization, abbreviated as Nptk共nasal # /p/, /t/, /k/兲 across intonational phrase, word, and syllable bound-aries. 共3兲 The third and final context elicited experimental data with a final nasal that allows for cross-boundary nasal-ization. The experimental data consisted of final nasals placed before voiced stops across intonational phrase, word, and syllable boundaries, abbreviated as Nblg共nasal # /b/, /l/, /+/兲. The comparison data in the Nptk context address the first research question involving the effect of boundary on final nasals, whereas the data in the Nblg context address the second research question concerning the strength of the in-tervening boundary on cross-boundary nasalization.
To minimize speaker-dependent articulatory variation, focus and declination were controlled in the corpus design. Although lexical stress does not exist in Taiwanese, words are produced with a longer duration and an expanded f0 range for lexical tones when under narrow focus共Pan, 2006兲. Therefore, the effect of focus condition is controlled by ask-ing the subjects to produce the utterances with a broad focus. To control declination effects, intonational phrase, word, and syllable boundaries were placed between the second and third syllables of sentences containing six syllables 关 #
兴. The final oral vowels and three final nasals, /m, n, G/ at the end of second syllable were followed by either one of the three initial voiced stops, /b, l, +/, or one of the three initial voiceless unaspirated stops, /p, t, k/, at the beginning of the third syllable,关VmnG兴 # 关bl+兴, 关VmnG兴 # 关ptk兴, 关V兴 # 关blg兴, or 关V兴 # 关ptk兴. Declination effect was minimized by having target syllables far from the end of the sentence.
Sentence 共1兲 is an example of a sentence used to elicit utterances with an intonational phrase boundary between the target syllables. In sentence共1兲, the first and second syllables formed a surname. A comma was placed after the second syllable in the sentence to elicit the production of the sur-name as a vocative. The right edge of the intonational phrase 共IP兲 boundary was defined by an f0 final lowering and a long pause and a postboundary pitch reset.
共1兲 In Taiwanese every lexical item has two tones, namely a juncture and a sandhi tone. A juncture tone surfaces on the last syllable within a tone group, whereas a sandhi tone
sur-faces on syllables located at the nonfinal position of a tone group. The domains of tone group are syntactically and pro-sodically determined; however, no complete account on Tai-wanese tone group delineation has been offered yet. In order to elicit productions with a word boundary but not with a tone-group boundary between the target syllables, sentences composed of an NP followed by an adjective phrase were used, as in共2兲.
共2兲 In sentence 共2兲, the first and second syllables, /si kin/ “four kilograms,” is a quantifier, whereas the third and fourth syllables /lai a/ “pear,” is the noun modifed by “four kilo-grams.” There is a word boundary between the second and third syllables, that is, between the quantifier and the noun. The tone-group boundary is located after the NP, e.g. “four kilograms of pears.”
To elicit utterances with a syllabic boundary between the target syllables, sentences such as 共3兲 containing a word spanning from the preboundary共second兲 syllable to the post-boundary 共third兲 syllable were used. The word containing target final nasal /oral vowel and initial stops was either a noun, or a verb, or an adjective. In the following case, the word was an adjective.
Tonal grouping:
共共sssj兲tone group共sj兲tone group兲IP Syntactic grouping:
关vp关v tshi G兴 关np关Adjkan tan e兴 关nhÅk tsÅG兴兴
wear simple clothing “Wear simple clothing . ”
共3兲
After designing the corpus, three reading lists were com-posed, with each reading list containing only sentences with only one type of boundary between target syllables. In other words, there was one reading list with sentences designed to elicit an intonational phrase boundary between second and third syllables, another to elicit a word boundary, and a third one to elicit a syllable boundary. The order of sentences within each list was randomized.
C. Instrumentation
Nasal airflow was recorded from a transducer mounted on a nasal airflow mask, a Hans Rudolph model P0789 adult nasal mask, which covered only the nose of the speaker. The transducer was connected to an MS100-A2 airflow system manufactured by Glottal Enterprises. Nasal airflow was
low-pass filtered at 36.5 Hz and then dc recordings were made by using a DT-2801 card installed in a PC. The signals were digitized at 11 kHz usingCSPEECHSPsoftware.
Acoustic signals picked up by a TEV microphone con-nected to a TEAC cassette deck were sent to a PC to be digitized simultaneously with nasal airflow at 11 kHz using CSPEECHSPsoftware.
A Seiko quartz metronome with blinking lights and tick-ing sounds to signal beats was used to regulate the global speaking rate. The ticking sounds were sent through a Bey-erdynamic headphone to pace speaking rates for the first and second syllables of each sentence. See below for further dis-cussion of how the global speaking rate was controlled.
D. Recording procedure
Airflow system calibration was carried out first by warming up the system for 1 h before recording and then adjusting the voltage displayed on the front panel of the sys-tem to zero “0” while no signals were being received. Then, the software was also calibrated by adjusting the volts of empty signals recorded to 0 volts as well. The flow was not actually calibrated; thus, the amplitudes reported in this study were volts rather than units of actual flow共e.g., cc/s兲. After hardware and software calibrations, the recordings were made in a sound-proofed room in the phonetics labora-tory of National Chiao Tung University, Taiwan. During the recordings, both the speaker and the experimenter were present in the sound-proofed room. The experimenter con-trolled the CSPEECHSPsoftware, signaled the speaker to put on the nasal airflow mask, and to read one sentence from the reading list. Speakers paused after each sentence to allow the experimenter time to save the nasal airflow and acoustic data at 11 kHz.
To control for global speaking rate, which previously had been found to influence nasal patterns, speakers wore a headphone emitting the ticking sounds from the metronome at a speed of 144 beats per minute. Based on native speakers’ judgments, this speed is compatible with the speaking rate during a normal conversation. None of the speakers reported that the speed deviated from what they normally used during conversation. There was also a blinking light on the metro-nome which was synchronized with the beats. Before the recordings were made, speakers were instructed to read the first and second syllables of the sentence at a speed with one ticking sound corresponding to the first syllable, and the sec-ond ticking sound correspsec-onding to the secsec-ond syllable. After pacing the speed at the beginning of the sentence, speakers were then free to slow down or speed up while producing the rest of the sentences to preserve the natural prosodic cadence of their utterances.
Altogether there were 162 target sentences recorded in the Nblg and Nptk contexts共3 final nasals ⫻6 initial conso-nants⫻3 prosodic boundaries ⫻3 repetitions兲 and 54 base-line sentences in the Vblg and Vptk contexts共1 final vowel ⫻6 initial consonants ⫻3 prosodic boundaries ⫻3 repeti-tions兲. Due to typing errors on the reading lists, three repeti-tions of one sentence that should have had a final /m/ fol-lowed by an /l/ across a syllable boundary and three
repetitions for each of the two sentences containing a final vowel followed by an /l/ across an intonational phrase boundary were discarded from the Nblg context.
E. Data analysis
As shown in Fig. 1, spectrograms generated by CSPEECHSPwere used to locate the time at the second vowel onsets 共V2On兲, and the offsets 共V2Off兲 as well as the third vowel onsets共V3On兲, and the offsets 共V3Off兲.
The time and amplitude of nasal airflow during produc-tion of the first nasal rise were taken to determine the timing and amplitude of nasality. These values were extracted at the following locations: 共1兲 the onset of the initial nasal rise 共On兲; 共2兲 the start of the first nasal plateau 共1On兲; 共3兲 the end of the first nasal plateau 共1Off兲; 共4兲 the offset 共Off兲 of the falling slope for the last nasal rise; and共5兲 the magnitude of the maximal nasal peak at the end of each utterance during normal exhalation was extracted共max兲 共Fig. 1兲. The start of a nasal plateau was the point in time when the rising slope of first nasal rise stopped and turned into a level contour, whereas the end of a nasal plateau was the point where the level nasal plateau started to change into a falling slope. If there was only a nasal peak without a plateau, then ampli-FIG. 1. Measurement points taken from nasal airflow with共a兲 two rises and 共b兲 one rise between the second and third vowels.
tudes and times of two points at the peak were taken. For nasal contours with two rises, the lowest point between the two rises共V兲 was also taken.
For the following discussion, refer to 共4a兲–共4e兲. After taking these measurements, further calculations were applied to these values to derive the temporal aspects of the nasal airflow, including the duration of the nasal plateau during the first nasal rise关R1PlDur 共4a兲兴, and the duration of the nasal airflow关NasDur 共4b兲兴. In utterances with two nasal rises, the duration of the nasal airflow did not truly represent the du-ration of the continuous nasal airflow because it included the duration of the nasal inhalation between the two nasal rises. According to Pan共2004兲, in the Nblg, Vblg, and Vptk con-texts, postboundary vowel nasalization occurs only during the initial 25% of vowel duration. In order to reveal the actual duration of nasalization during postboundary initial stops and vowels, the current study calculated the offset la-tency of nasal airflow relative to the onset of the third vowel, by subtracting the time at the offset of the nasal airflow from the time of the onset of the third vowel, as shown in 关CV-NasDur 共4c兲兴. The positive values of the duration of post-boundary initial consonant and vowel nasalization 共CVNas-Dur兲 indicate the duration of the third vowel nasalization, whereas the negative values revealed the duration of nasal-ization in the initial voiced and voiceless stops. In addition to the magnitude and temporal aspect of nasal airflow, the du-ration of the second and third vowel /a/ was also calculated in关V3Dur 共4d兲; V2Dur 共4e兲兴.
Duration of first nasal plateau共R1PlDur兲 = time at end of first nasal plateau共1Off兲
- time at start of first nasal plateau共1On兲; 共4a兲 Duration of nasal airflow共NasDur兲
= time at offset of nasal airflow共Off兲
- time at onset of first nasal airflow rise共On兲; 共4b兲 Duration of postboundary stop and vowel nasalization
共CVNasDur兲
= time at offset of nasal airflow共Off兲
- time at onset of third vowel共V3On兲; 共4c兲 Duration of postboundary vowel /a/共V3Dur兲
= offset of postboundary vowel共V3Off兲
- onset of postboundary vowel共V3On兲; 共4d兲 Duration of preboundary vowel /a/共V2Dur兲
= offset of preboundary vowel共V2Off兲
- onset of pre-boundary vowel共V2On兲. 共4e兲 The overall nasal amplitude contour, the nasal ampli-tudes at the start and end of the first nasal plateau, and the derived parameters were analyzed statistically. The purpose
of the statistical analysis was to determine interactions using three-way repeated analysis, to isolate the interactions by dividing the data into subsets according to interactions, to reveal the effect of boundary on each data set, and to identify the ranking and distinction for levels of boundary. The ␣ level was set at 0.05. The following is the rationale behind each statistical analysis.
Due to the widely reported speaker-dependent variations in previous literature, even though speaker and context were not of interest in the present study, their effects must be determined before data can be pooled across different factors for further statistical analyses. Thus, a three-way repeated MANOVA共speaker⫻context⫻boundary兲 was used to ana-lyze the effect of speaker共speaker 1, 2, and 3兲 and contexts 共Vblg, Vptk, Nblg, and Nptk兲 and boundaries 共intonational phrase, word, and syllable兲 on six dependent variables, in-cluding nasal amplitudes at the offset of the second vowel, the onset of nasal airflow, the start of the first nasal plateau, the end of first nasal plateau, the offset of nasal airflow, and the onset of the third vowel. For utterances that showed a flat contour of nasal airflow, the average nasal amplitude at the offset of the second vowel and the onset of the third vowel were used to replace the values at the onset of the first nasal rise, the start of the nasal plateau during the first nasal rise, the end of the first nasal plateau, and the offset of nasal airflow. It should be noted that the repeated measurements were actually the six data points from each of the three rep-etitions of the nine sentences in the Nblg or Nptk contexts, and the three sentences in the Vblg or Vptk contexts.
After determining the interactions with the three-way MANOVAs, the data were further divided according to con-texts, Nblg, Nptk, Vblg, and Vptk, if there was a two-way interaction between context and boundary, but no three-way interactions between speaker, context, and boundary. For pa-rameters with significant three-way interactions between speaker, context, and boundary, the data were further divided into 12 different subsets according to the three speakers and four contexts, speaker 1 Nblg, speaker 1 Nptk, speaker 1 Vblg, speaker 1 Vptk, speaker 2 Nblg, speaker 2 Nptk, speaker 2 Vblg, speaker 2 Vptk, speaker 3 Nblg, speaker 3 Nptk, speaker 3 Vblg, and speaker 3 Vptk. After dividing the data into subsets according to interactions, one-way MANO-VAs共boundary兲 on each of the six previously discussed sub-data sets were used to explore the effect of boundary on overall nasal contours.
In addition to the overall nasal contour, the nasal ampli-tudes at specific points, including the start and end of the first nasal plateau共1On, 1Off兲, were divided into subsets accord-ing to the interactions on nasal contours, and then analyzed with one-way repeated ANOVAs 共boundary兲 and posthoc Duncan tests to access the possibility of boundary distinction at particular time points.
The four parameters derived from further calculations, i.e. 共4a兲 to 共4d兲, were each analyzed with a three-way re-peated ANOVA共speaker⫻context⫻boundary兲 to determine the nature of the interactions. Data for each of the four pa-rameters showing interactions between context and bound-ary, but no three-way interactions between speaker, context, and boundary, were further divided into four data sets
ac-cording to contexts. The data of a parameter showing three-way interactions between speaker, context, and boundary were further divided into 12 data sets according to three speakers and four contexts. Each of these data sets was ana-lyzed with a one-way repeated ANOVA共boundary兲 to access the effect of boundary. It should be noted that the repeated measurements were theoretically the 216 repetitions 共3 speakers⫻3 boundaries ⫻3 final nasals ⫻3 stops ⫻3 rep-etitions兲 in the Nblg and Nptk contexts, and the 81 repeti-tions 共3 speakers ⫻3 boundaries ⫻1 final vowel ⫻3 stops ⫻3 repetitions兲 in the Vblg and Vptk contexts. However, since the first nasal rise did not appear in some repetitions in the Nblg and Nptk contexts or in most of the repetitions in the Vblg and Vptk contexts, the actual number of observa-tions varied among the parameters. After analyzing the main effect of boundary, posthoc Duncan tests were performed to reveal the ranking and differences in the levels of boundary for each derived parameter.
III. RESULTS
A. Global speaking rate
This study controlled the global speaking rate by pacing the speakers’ production of the first and second syllables at a speed of 144 beats per minute. To access the effectiveness of this speed-controlling method, the duration of the second vowel /a/ was compared between three speakers. The means and standard deviations of /a/ produced by speaker 1 were
longer than those by speakers 2 and 3 共speaker 1: mean = 159 ms, S . D . = 47 ms; speaker 2: mean= 148 ms, S . D . = 42 ms; speaker 3: mean= 146 ms, S . D . = 50 ms兲. The re-sults of 2-tailed t-tests showed that /a/ was significantly longer in productions by speaker 1 than by speaker 2 共p ⬍0.01兲 or by speaker 3 共p⬍0.01兲. However, the duration of /a/ produced by speakers 2 and 3 were not significantly dif-ferent from each other共p=0.67兲. Not all speakers responded to the metronome method in a similar pattern; speaker 1 exhibited a slower rate and produced longer vowels than were exhibited by speakers 2 or 3. Perceptually, speaker 1 indeed sounded slower and softer than speakers 2 and 3.
B. Nasal contour patterns and boundary types
As illustrated in Fig. 2, excluding the data from speakers 2 and 3 in the Nblg and Nptk contexts, the amplitudes of nasal airflow during the utterances were lower than the maxi-mal nasal exhalation peak at the end of utterances 共max兲. Therefore, speakers only used a portion of the nasal ampli-tude range that was available to them in most productions.
The average nasal contours across the intonational phrase boundary exhibited a first nasal rise followed by a nasal valley共inhalation兲 and then a second nasal rise. Across the word boundary in the Nblg and Nptk contexts, the only pattern of nasal contours observed was one nasal rise, but in the Vblg and Vptk contexts, only flat nasal contours could be FIG. 2. Amplitudes of nasal airflow taken at ten time points through the production of target segments by three speakers at three boundaries types共intonational phrase, word, and syllable兲 in four segmental contexts 共Nblg: final nasal followed by initial voiced stop, Nptk: final nasal followed by initial voiceless stop, Vblg: final oral vowel followed by initial voiced stop, Nptk: final nasal followed by initial voiceless stop兲. The points include: offset of second vowel 共V2Off兲, onset of nasal airflow共On兲, start of the first nasal plateau 共1On兲, end of the first nasal plateau 共1Off兲, nasal inhalation 共V兲, start of the second nasal plateau 共2On兲, end of the second nasal plateau 共2Off兲, offset of nasal airflow 共Off兲, onset of the third vowel 共V3On兲, and maximal nasal exhalation at the end of utterance共Max兲.
observed. Nasal contours across a syllable boundary exhib-ited either one nasal rise or a flat nasal contour in all four contexts.
Detailed observation of the nasal contours revealed five patterns: 共1兲 a first nasal rise followed by nasal inhalation, seen as a nasal valley with a negative nasal amplitude, and then a second nasal rise共RVR兲, as shown in Fig. 1共a兲; 共2兲 a nasal rise共R兲, as shown in Fig. 1共b兲; 共3兲 a nasal rise followed by a nasal inhalation共RV兲, as shown in Fig. 3共a兲; 共4兲 a nasal inhalation followed by a nasal rise 共VR兲, as shown in Fig. 3共b兲; and 共5兲 no nasal rise or inhalation 共flat兲, as shown in Fig. 3共c兲. The distribution of the five nasal patterns across speakers and boundary types is shown in Table I.
As shown in Table I, nasal inhalation could only be ob-served across an intonational phrase boundary when there was a pause between intonational phrase boundaries; more-over, second nasal rises were observed only across intona-tional phrase boundaries. Although nasal inhalation and a second nasal rise occurred only across intonational phrase boundaries, not all IPs were marked in this manner. Gener-ally speaking, nasal inhalation occurred more often in the
Nblg and Nptk contexts than in the Vblg and Vptk contexts. The frequency of nasal inhalation across intonational phrase boundaries varied from speaker to speaker. The presence of a nasal rise and nasal inhalation across an intonational phrase boundary in the Vblg and Vptk contexts where no nasal seg-ments were present suggested that the intonational phrase boundary was a strong factor in inducing velic movements.
Across a word boundary in the Nblg and Nptk contexts, the pattern of nasal contours was predominantly a single na-sal rise共Fig. 2兲. The exceptions were one utterance produced by speaker 1 with VR pattern and five utterances produced by speaker 3 with flat nasal contours. In the Vblg and Vptk contexts, the nasal contours were flat across all speakers 共Table I兲.
Across the syllable boundary in the Nblg and Nptk con-texts, the nasal patterns were one nasal rise for all, except for one production by speaker 1 in the Nblg context. In the Vblg and Vptk contexts, a flat nasal contour was the most com-monly observed pattern.
In short, the existence of a strong prosodic boundary stopped nasalization from extending into the neighboring segment and so speakers produced two separate nasal con-tours. In contrast, when there was a weaker syllable or word boundary, nasalization extended over the boundary and so speakers were likely to produce only one nasal rise.
C. Nasal amplitude and boundary type
Nasal amplitudes taken at six points during nasal con-tours, including the offset of the second vowel, the onset of the nasal rise, the start of the first nasal plateau, the end of the first nasal plateau, the offset of the nasal airflow, and the onset of the third vowels, were analyzed with a three-way MANOVA 共speaker⫻context⫻boundary兲. The results of a three-way MANOVA revealed a significant three-way
inter-action between speaker, context, and boundary
关F共72,3096.1兲=3.53, p⬍0.0001兴. In other words, the effect of boundary on nasal amplitudes produced by the three speakers varied according to context. Thus, it was not pos-sible to pool the data either across speakers or contexts. As a result, the data were further divided into 12 data sets accord-ing to the three speakers and four contexts.
To further determine the effect of boundary on overall nasal contours, 12 one-way repeated MANOVAs共boundary兲 were used to analyze the nasal contours in the Nblg, Nptk, Vblg, and Vptk contexts that were produced by all the speak-ers. The results of the 12 one-way repeated MANOVAs 共boundary兲 are shown in Table II. The significant effects of boundary can be observed in the nasal contours produced by the three speakers in the four contexts. In other words, the nasal contours were influenced by the boundary types, re-gardless of the contexts and speakers.
In addition to the nasal contours, the nasal amplitude at specific points, including the start and the end of the first nasal plateau共1On, 1Off兲, were also analyzed. As the results of previous three-way repeated MANOVAs have shown three-way interactions between speakers, contexts, and boundaries, the nasal amplitudes at the start of the first nasal plateaus were divided into 12 sets according to the three FIG. 3. Examples for nasal contours with共a兲 a nasal rise followed by a
valley共RV兲; 共b兲 a valley followed by a nasal rise 共VR兲; and 共c兲 no nasal rise or valley共flat兲. Nasal airflow is aligned with the speech signal.
TABLE I. Nasal patterns categorized according to boundaries and contexts. “I,” intonational phrase boundary; “W,” word boundary; “S,” syllable boundary; RVR, first nasal rise followed by a nasal valley共inhalation兲, and then a second nasal rise; RV, first nasal rise followed by a nasal valley 共inhalation兲; VR, a nasal valley 共inhalation兲 followed by a nasal rise; R, one nasal rise; Flat, no nasal rise or valley.
Speaker 1 Nblg Nptk Vblg Vptk I W S I W S I W S I W S RVR 17 7 RV 1 1 VR 1 3 3 R 7 26 23 19 27 24 5 2 6 2 Flat 1 9 7 9 7 Speaker 2 Nblg Nptk Vblg Vptk I W S I W S I W S I W S RVR 4 3 2 RV VR 1 R 21 27 24 24 27 24 7 1 7 Flat 9 8 9 9 Speaker 3 Nblg Nptk Vblg Vptk I W S I W S I W S I W S RVR 10 8 3 4 RV 1 4 2 2 VR R 14 25 24 15 24 24 4 2 3 Flat 2 3 9 6 9 9
speakers and four contexts, as were the nasal amplitudes at the end of the first nasal plateau. Each of the data sets was then analyzed with a one-way repeated ANOVA共boundary兲. The results of the one-way ANOVAs 共boundary兲 revealed a significant effect of boundary in almost all data sets, exclud-ing the production of speakers 2 and 3 in the Nptk context 共Table III兲. Furthermore, the posthoc Duncan tests revealed
no consistent cross-speaker ranking of boundary type in the Nblg and Nptk contexts 共Table III兲. In other words, in the Nblg and Nptk contexts, the nasal amplitudes at the start and the end of the first nasal plateaus were not used to mark boundary types. However, in the Vblg and Vptk contexts there was a consistent ranking of boundary types, intona-tional phrase⬎syllable⬎word, on nasal amplitudes at the TABLE II. Results of one-way repeated MANOVAs共boundary兲 on nasal contours produced by three speakers
in four contexts. “**” p⬍0.01.
Nblg Nptk Vblg Vptk
Speaker 1 F共12,132兲=19.35** F共12,140兲=48.81** F共12,32兲=8.12** F共12,32兲=24.62**
2 F共12,136兲=15.94** F共12,140兲=16.03** F共12,38兲=22.45** F共12,36兲=11.53**
3 F共12,132兲=16.06** F共12,134兲=13.18** F共12,36兲=18.93** F共12,38兲=20.13**
TABLE III. One-way ANOVA共boundary兲 and posthoc Duncan tests on amplitudes of nasal airflow at onset 共1On兲 and offset 共1Off兲 of the first nasal plateau. The intonational phrase boundary is represented by “I,” the word boundary by “W,” the syllable boundary by “S.” “=” the levels of boundary were not significantly different from each other. “⬎” the levels of boundary were significantly different from each other. “*” p
⬍0.05; “**” p⬍0.01.
1On 1Off
ANOVA ANOVA
Context Speaker Boundary Mean posthoc Duncan Mean posthoc Duncan
Nblg 1 I 0.412 共2,71兲=4.12* 0.532 共2,71兲=3.15* W 0.359 S⬎I=W 0.616 S = W, W = I S 0.544 0.708 S⬎I 2 I 1.277 共2,73兲=31.57** I⬎S=W 0.939 共2,73兲=5.51** I⬎S=W W 0.564 0.640 S 0.737 0.919 3 I 1.790 共2,71兲=5.95** I = W, W = S I⬎S 2.142 共2,71兲=8.21** I⬎W=S W 1.467 1.571 S 1.081 1.094 Nptk 1 I 0.393 共2,75兲=25.94** S⬎W⬎I 0.771 共2,75兲=8.04** S⬎W=I W 0.578 0.903 S 0.830 1.083 2 I 1.303 共2,75兲=7.72** I = S⬎W 1.271 共2,75兲=0.74, p = 0.48 W 0.852 1.178 S 1.088 1.273 3 I 1.976 共2,72兲=1.44 p = 0.24 2.548 共2,72兲=2.03 p = 0.14 W 1.695 2.201 S 2.023 2.171 Vblg 1 I 0.478 共2,21兲=11.02** I = S⬎W 0.541 共2,21兲=11.20** I = S⬎W W −0.039 −0.039 S 0.380 0.465 2 I 0.444 共2,24兲=28.47** I⬎S=W 0.550 共2,24兲=28.11** I⬎S=W W −0.090 −0.090 S 0.043 0.044 3 I 0.855 共2,23兲=13.37** I⬎S⬎W 0.967 共2,23兲=18.79** I⬎S⬎W W −0.238 −0.238 S 0.229 0.251 Vptk 1 I 1.517 共2,21兲=121.88** I⬎S⬎W 0.294 共2,21兲=12.48** I = S⬎W W −0.063 -0.063 S 0.260 0.193 2 I 0.795 共2,23兲=21.35** I⬎S=W 1.033 共2,23兲=31.46** I⬎S=W W −0.096 −0.096 S 0.009 0.004 3 I 0.740 共2,24兲=19.93** I⬎S=W 0.803 共2,24兲=34.67** I⬎S=W W −0.256 −0.256 S −0.082 −0.082
start and the end of the first nasal plateau. By comparing the distribution of nasal contour patterns in Table I and the mean nasal amplitudes in Table III, it was found that this consistent ranking in the Vblg and Vptk contexts was due to the con-stant surface of the first nasal rise around an intonational phrase boundary which contributed to the high mean nasal amplitudes, the occasional surface of the first nasal rise at syllable boundary which led to the second highest mean na-sal amplitude, and the lack of a first nana-sal rise at a word boundary which led to negative nasal amplitudes. In other words, speakers rarely produce a nasal rise in the Vblg and Vptk contexts. Even if they produced nasal rises, as in the Nptk and Nblg contexts, the nasal amplitudes at the start and the end of the first nasal plateau did not consistently vary with hierarchical boundary influences.
D. Temporal domain of nasal airflow and boundary type
Turning to the temporal domain of nasal airflow, as shown in Fig. 4, after the start of the first nasal plateau, as the
nasal rises and valleys began to emerge, the duration of nasal airflow increased at a faster rate at the intonational phrase boundary than at other boundaries. Moreover, the duration of the nasal airflow was found to be longer at an intonational phrase boundary than at word and syllable boundaries. The temporal domain of a nasal airflow remained relatively the same at word and syllable boundaries.
As shown in Fig. 4, the duration of the first nasal plateau 关R1PlDur, 共4a兲兴, between the start 共1On兲 and end 共1Off兲 of the first nasal rise, was shorter at word or syllable boundaries than at intonational phrase boundary. In response to the sig-nificant two-way interaction between context and boundary on the duration of the first nasal plateau revealed by a three-way repeated ANOVAs 共speaker⫻context⫻boundary兲 共Table IV兲, the data were further divided into four sets ac-cording to contexts to isolate the interactions. After dividing the data, the effect of boundary on each subdata set was further analyzed through one-way ANOVAs 共boundary兲. As shown in Table V, the results of four one-way repeated FIG. 4. Six latency measures共in seconds兲 of nasality taken during the production target sounds. Measures include: the end of the second vowel 共V2Off兲, the onset of the nasal airlfow共On兲, start of the first nasal plateau 共1On兲, end of the first nasal plateau 共1Off兲, the offset of nasal airflow 共Off兲, and the onset of the third vowel共V3On兲 relative to offset of the second vowel 共V2Off兲.
TABLE IV. Significant interactions according to results of five three-way repeated ANOVAs 共subject ⫻context⫻boundary兲 on duration of first nasal plateau 共R1PlDur兲, duration of nasal airflow 共NasDur兲, duration of postboundary stop and vowel nasalization共CVNasDur兲, and duration of the third vowel 共V3Dur兲.␣= 0.05.
Parameters N Interactions F value p value
R1PlDur 504 Context⫻boundary F共4,476兲=3.59 p⬍0.01
subject⫻context⫻boundary F共6,476兲=0.38 p = 0.89
NasDur 514 Context⫻boundary F共4,486兲=11.03 p⬍0.01
subject⫻context⫻boundary F共6,486兲=1.32 p = 0.25
CVNasDur 515 Context⫻boundary F共4,487兲=2.99 p⬍0.05
subject⫻context⫻boundary F共6,487兲=1.28 p = 0.26
V3Dur 570 Context⫻boundary F共6,534兲=5.95 p⬍0.01
ANOVAs共boundary兲 revealed a significant effect of bound-ary type on the duration of the first nasal plateau共R1PlDur兲 in the Nblg and Nptk contexts. A posthoc Duncan test showed the ranking for the mean duration of the first nasal plateau in the Nblg and Nptk contexts: intonational phrase ⬎word⬎syllable. However, the durations of the first nasal plateau at word and syllable boundaries in the Nblg contexts were grouped together.
In the Vblg and Vptk contexts, due to the lack of a first nasal rise at the word boundary and a smaller number of tokens with a first nasal rise at the syllable boundary共Table I兲, there were no consistent rankings of boundary types on the duration of nasal plateau in the Vblg and Vptk contexts. In other words, only in the Nblg and Nptk contexts was the duration of the first nasal rise significantly longer at the IP than at the word and syllable boundaries.
As shown in Fig. 4, the duration of nasal airflow 关Nas-Dur 共4b兲兴, between the onset of nasal airflow 共On兲 and the offset of nasal airflow 共Off兲, was longer at an intonational phrase than at word and syllable boundaries. Results of
three-way repeated ANOVAs 共speaker⫻context
⫻boundary兲 revealed significant two-way interactions
be-tween context and boundary affecting the duration of nasal airflow共NasDur兲 共Table IV兲. In response to this observation, the data were further divided into four sets according to con-texts.
Results of the four one-way repeated ANOVAs 共bound-ary兲 and the posthoc Duncan tests revealed significant effects of boundary on nasal airflow duration共NasDur兲 in the Nblg, Nptk, and Vblg contexts. There were consistent cross-boundary rankings on the mean duration of nasal airflow 共NasDur兲, intonational phrase⬎word艌syllable 共Table V兲. However, the posthoc Duncan tests tended to group nasal duration at the word and syllable boundaries together in the Nblg and Nptk contexts. In other words, the duration of nasal airflow was longer at an intonational phrase boundary than at word and syllable boundaries.
In sum, the intonational phrase boundary was marked by the longest duration of the first nasal plateau and nasal air-flow. Although the duration of the first nasal plateau and nasal airflow at a word boundary was the second longest, they were not significantly longer than the duration of the first nasal plateau and nasal airflow at a syllable boundary. TABLE V. Results of one-way repeated ANOVA共boundary兲 and posthoc Duncan tests on the duration of the
first nasal plateau共R1PlDur兲, the nasal airflow 共NasDur兲, the postboundary stop and vowel nasalization 共CV-NasDur兲, and the postboundary vowel 共V3Dur兲. “I” intonational phrase boundary, “W” word boundary, “S” syllable boundary. “=” the levels of boundary were not significantly different from each other. “⬎” the levels of boundary were significantly different from each other “**” p⬍0.001, “*” p⬍0.05.
R1PlDur NasDur
Context Boundary Means
ANOVA
posthoc Duncan Means
ANOVA posthoc Duncan Nblg I 0.167 共2,221兲=29.89** I⬎W=S 0.447 共2,222兲=247.13** I⬎W=S W 0.089 0.211 S 0.084 0.196 Nptk I 0.124 共2,228兲=70.56** I⬎W⬎S 0.333 共2,228兲=201.84** I⬎W=S W 0.052 0.163 S 0.035 0.147 Vblg I 0.075 共1,26兲=1.6 p = 0.217 0.368 共1,29兲=13.57** I⬎S W S 0.026 0.084 Vptk I 0.055 共1,19兲=13.36** S⬎I 0.262 共1,25兲=0.01 p = 0.917 W S 0.195 0.272 CVNasDur V3Dur
Context Boundary Means
ANOVA
posthoc Duncan Means
ANOVA posthoc Duncan Nblg I −0.001 共2,221兲=9.26** S = W⬎I 0.144 共2,208兲=43.10** W⬎S⬎I W 0.018 0.202 S 0.019 0.163 Nptk I −0.118 共2,228兲=52.03** W = S⬎I 0.171 共2,231兲=46.68** W⬎I⬎S W −0.049 0.218 S −0.056 0.155 Vblg I −0.084 共1,30兲=0.48 p = 0.494 0.160 共2,41兲=8.19** W⬎S=I W 0.243 S 0.021 0.175 Vptk I −0.119 共1,26兲=7.42* S⬎I 0.164 共2,78兲=20.37** W⬎S⬎I W 0.244 S 0.018 0.206
E. Postboundary initial stop and vowel nasalization
Turning to the duration of the postboundary stop and vowel nasalization 共CVNasDur兲, which was determined by the offset of nasal airflow relative to the onset of the post-boundary vowel, it can be seen in Fig. 5 that there was a trend for nasal airflow to cease before the onset of the post-boundary vowel in the Nptk and Vptk contexts, but after the onset of the postboundary vowel in the Nblg and Vblg con-texts. In addition, boundary type also played a role in the extent of nasalization. Cross-boundary nasalization in these former two contexts was blocked earlier relative to the onset of the third vowel when there was an intonational phrase boundary. In other words, nasalization carried over from one word to the next in the Nblg and Vblg contexts as long as
there was no intervening intonational phrase boundary. There were no data at word and syllable boundaries in the Vblg and Vptk contexts, as shown in Fig. 5. This was because the nasal contour remained flat in these conditions, as shown in Table I; therefore no nasalization was observed.
Due to a significant two-way interaction between con-text and boundary on duration of nasalization for initial stops and the vowel in the postboundary position 共Table IV兲, the data were further divided into four sets according to con-texts. The results of the one-way repeated ANOVAs 共bound-ary兲 show a significant effect of boundary in the Nblg, Nptk, and Vptk contexts. The only consistent ranking observed was that the nasalization was shortest when an intonational phrase boundary separated words共Table V兲.
To sum up, as shown in Fig. 5, in the Vptk context there FIG. 5. Box plots for latency between the offset of nasal airflow共Off兲 and the onset of the third vowel 共V3On兲 produced by three speakers, in four contexts 共Nblg: the final nasal followed by the initial voiced stop, Nptk: the final nasal followed by the initial voiceless stop, Vblg: the final vowel followed by the initial voiced stop, and Vptk: the final vowel followed by the initial voiceless stop兲 across three boundaries 共IP: intonation phrase, WD: word, and SYL: syllable boundaries兲. The means are indicated by the lines in the middle of the boxes, standard deviations are indicated by the frame of boxes, and the 10% and 90% data ranges are indicated by the whiskers.
was no nasalization. In the Vblg context, the nasality termi-nated after the IP boundary. In the Nptk context, the nasality was terminated before or during postboundary initial voice-less stops. In the Nblg context, the nasalization process spread from the final nasal across the boundary into the post-boundary initial voiced stop and vowel. Moreover, the nasal airflow was terminated earlier across an IP boundary than across other boundaries.
F. Vowel length
The results of a three-way repeated ANOVAs 共speaker ⫻context⫻boundary兲 found that there was a significant two-way interaction between context and boundary on the length of the postboundary vowel /a/ 关V3Dur 共4d兲兴 共Table IV兲. Therefore, the durations of the third vowel /a/ were di-vided according to context: Nblg, Nptk, Vblg, and Vptk. The results of the four one-way repeated ANOVAs 共boundary兲 and the Duncan tests revealed that the duration of the post-boundary vowel was consistently the longest after a word boundary in all contexts共Table V兲. In other words, the long-est postboundary vowel duration was observed after word boundary.
IV. DISCUSSION
A. Nasality and boundary type
This study addresses two research questions, namely共1兲 the effect of boundary on nasality, and 共2兲 the effect of boundary on cross-boundary nasalization. The effect of boundary on nasality was revealed through patterns of nasal contours, the duration of the first nasal plateau, the duration of nasal airflow, and the measurement of nasal temporal la-tency and nasal amplitudes at the onset of nasal airflow, the start and the end of the first nasal plateau, and the offset of nasal airflow. The effect of boundary on nasalization was revealed by the duration of nasalization on initial stop and following vowel combinations in postboundary syllables.
It was found that boundary influences the pattern of na-sal contour. For example, nana-sal patterns with a first nana-sal rise followed by a nasal inhalation and a second nasal rise can only be observed at an intonational phrase boundary, whereas nasal contours with one nasal rise were observed at word and syllable boundaries in the Nblg and Nptk contexts. Flat nasal contours were found at word and syllable bound-aries in the Vblg and Vptk contexts.
In order to capture the differences in patterns of nasal contour, various measures, including 共1兲 magnitude differ-ences and共2兲 temporal aspects of nasal airflow, were inves-tigated. Although there were significant effects of boundary on overall nasal contours, neither the nasal amplitudes at the start共1On兲 or the nasal amplitudes at the end 共1Off兲 of the first nasal rise varied consistently according to boundary type in the Nblg and Nptk contexts. Regardless of the lack of a consistent pattern in the magnitude of nasal airflow, the du-ration of the first nasal plateau共NasPlDur兲 and nasal duration 共NasDur兲 varied consistently according to boundary type in the Nblg and Nptk contexts: intonational phrase⬎word 艌syllable. It was observed that the stronger the prosodic boundary, the longer was the nasal duration and the duration
of the first nasal plateau in both the Nptk and Nblg contexts. It should be noted that the durations of the first nasal plateau and nasal airflow at word and syllable boundary tend to be grouped together in the current study. It is actually a com-mon finding in articulatory prosody studies that word and syllable boundaries are not distinguished. Future studies are necessary to further explore this issue.
With reference to the first research question identified earlier regarding the influence of prosodic boundary on na-sality, it can be said that although there were no consistent patterns in the magnitude of nasalization at the onset and offset of the first nasal plateau, due to the different amplitude ranges used by different speakers, each speaker realized boundary strength differences in the same direction共Fig. 2兲. In addition, there was an effect of boundary strength on the overall nasal trace for individual speakers 共Table II兲. More-over, there were consistent patterns in the temporal domain that varied in relation to boundary type. These included the duration of nasal airflow and duration of the first nasal pla-teau.
Taiwanese is not the only language where prosodic in-fluences on articulation are more prominent on temporal pat-terns than on spatial magnitude patpat-terns. For example, Byrd et al. 共2000兲 in a study of the jaw, lip, and lingual move-ments of 关n#m兴 and 关m#n兴 across the word and phrasal boundaries in Tamil, discovered little effect of phrasal posi-tion on the spatial domain, but did find a significant effect of phrasal position on timing and duration. In the current study boundaries influence the pattern of nasal contour and nasal temporal, but not nasal magnitude, parameters in Taiwanese. The lack of consistent variations in nasal magnitudes at the onset and offset of the nasal plateau across speakers may be random variability, due to the fact that once the velopha-ryngeal port is open, it is difficult to control airflow magni-tude in a sufficiently variable manner, especially if respira-tory effort remains fairly constant. Thus, nasal magnitudes do not vary according to boundary type. Alternatively, it is suggested that there may be several patterns of boundary effects on nasal magnitudes that speakers can choose from; thus, no consistent rankings of boundary type can be ob-served. To fully reveal the patterns of nasal magnitudes that speakers can use, in further prosodic articulatory studies, more subjects are needed in order to find out how many patterns of variations actually exist.
Although the results of this study indicate that nasal magnitude does not vary according to boundary type in the Nblg and Nptk contexts, the variation in the duration of nasal airflow in relation to hierarchical boundary strength suggests that speakers may use temporal cues to signal boundary types. In fact, a study which did not use speech signals as stimuli found that for two tones played with the same fre-quency but with differing lengths, the thresholds for longer signals were lower in dB than for short signals共Watson and Gengel, 1969; Yost, 2000兲. In other words, the longer signals required less intensity to be perceived, whereas the shorter signals required higher intensity. Thus, the duration of a sound can influence its perceptibility. Furthermore, Ha et al. 共2004兲 found that the mildly hypernasal speech of children with a cleft palate showed longer temporal characteristics of
a nasal onset interval, nasal offset interval, and total nasal-ization duration, than did children of the same age without cleft palates. In other words, the longer nasal temporal char-acteristics may contribute to the perception of hypernasality. This evidence suggests that nasal duration affects the degree of perceived nasality. Further perceptual studies are neces-sary to explore the relationship between nasal duration and nasality as perceived at different boundary types.
B. Cross-boundary nasalization and boundary type
With reference to the second research question regarding the effect of boundary on cross-boundary nasalization, this study has found that cross-boundary nasalization terminates earlier when it occurs across an intonational phrase bound-ary. This pattern supports Cho’s hypothesis 共2004兲 which states that the stronger the prosodic boundary, the more re-sistance there is to cross-boundary coarticulation. Further-more, as articulatory gestures become more canonical, as when around a stronger boundary, the gesture is less likely to coarticulate with the neighboring segments. Both the hierar-chical influence of boundary on domain-edge strengthening and the effect of intervening boundary on cross-boundary coarticulation find support in this paper.
When considering context, it was found that nasal air-flow ceases before the end of words closed with a voiceless stop共Nptk context兲, but continues after a word boundary in words closed with a voiced stop 共Nblg context兲. In other words, nasalization can cross over voiced consonants. It is proposed that there are speech aerodynamic reasons behind the different nasalization patterns observed in Nptk and Nblg contexts. First, from the point of building up intraoral air pressure, complete vocal-tract closure involving the oral cav-ity and the velopharyngeal port is necessary for voiceless stops to develop and maintain the required intraoral air pres-sure between the oral cavity and the atmosphere. Second, in reference to the maintenance of voicing, the opening of the velopharyngeal port during the production of voiced stops helps to release the supralaryngeal pressure, which must be lower than the sublaryngeal pressure, in order for pulmonic air to flow through the glottis and to cause the vocal folds to vibrate共Ohala and Riodan 1979兲. Thus, for aerodynamic rea-sons, the closure of the velopharyngeal port leads to the ces-sation of nasality during voiceless stops, and the opening of the velopharyngeal port leads to nasalization during voiced stops.
C. Idiosyncratic and gradient hierarchical boundary markers
In addition to the prosodic and hierarchical influence of boundaries on patterns of nasal contours and nasal temporal parameters关i.e., duration of the first nasal plateau 共R1PlDur兲 and duration of the nasal airflow共NasDur兲兴, there are acous-tic markers uniquely associated with each boundary type that facilitate boundary identification. For example, an intona-tional phrase boundary can be marked by nasal inhalation and a second nasal rise, the longest duration of a first nasal plateau and duration of nasal airflow, or the shortest duration of postboundary stops and vowel nasalization共CVNasDur兲,
whereas a syllable boundary is marked by the shortest dura-tion of nasal temporal parameters, such as duradura-tion of the first nasal plateau and duration of nasal airflow. As for a word boundary, it is marked by the longest duration of the postboundary vowel 共V3Dur兲. In fact, this same effect was also discovered in Pan’s共in press兲 study on the initial lexical tone after a word boundary. As there are two types of word boundaries in Taiwanese, it is proposed that the longest post-boundary vowel duration marks a word post-boundary that does not coincide with a tone group boundary, whereas the surface of juncture tone in a preboundary vowel marks a word boundary that coincides with a tone group boundary. Future studies are necessary to further explore this issue.
Not all articulatory or acoustical cues contribute equally to perception. Wightman et al.共1992兲 in a perceptual study of the boundary effect on final lengthening found that the lengthening of the rhyme of the syllables preceding bound-aries can be used perceptually to distinguish at least four types of boundaries. They asked subjects to assign seven levels of break indices to a read speech corpus. The results showed that different boundaries were distinguished by dif-ferent sets of durational cues including the coda consonants, the vowel nucleus, all segments between the final stressed vowel and the final vowel, and the final stressed vowel of preboundary syllables. Moreover, the perception of certain boundaries relied more on the duration of coda consonants and vowel nuclei, whereas the perception of other boundary types relied more on the duration of the final stressed vowels. In the present study, different types of boundaries have been distinguished hierarchically not only by nasal temporal cues, but also by unique phonetic markers, such as patterns of the nasal contour and the duration of the postboundary vowel.
From a perceptual point of view, further perceptual ex-periments which analyze the perceptual salience of patterns of nasal contour, hierarchical nasal temporal cues, and idio-syncratic boundary markers can shed light on cues that lis-teners use to distinguish boundary types. As the influence of language-specific phonemic categorization has been demon-strated to influence the perception of nasality and cross-boundary nasalization 共Beddor and Strange, 1982; Beddor and Krakow, 1999兲, it is necessary to conduct cross-linguistic perceptual studies to enhance our understanding of domain-edge nasality strengthening and cross-boundary nasalization.
V. CONCLUSIONS
The present study reveals how boundary type influences the pattern of nasal contours and the temporal aspects of both acoustic and nasal airflow data in Taiwanese. In addition to the hierarchical ranking of boundary type on nasal temporal parameters, each boundary type is associated with specific phonetic markings that facilitate the identification of bound-ary. Finally, with a view to future work in this area, it would be interesting to expand the scope of prosodic articulatory studies on nasality into the area of speech pathology and to explore the effect of boundary on nasality and nasalization produced by subjects who exhibit hypernasal speech.
