Third TIE Conference on Tone and Intonation –University of Lisbon - September 15-17 2008University of Lisbon September 15 17, 2008

Intonational phrase boundaries in BP Intonational phrase boundaries in BP ppacross speech styles: across speech styles:

perception and productionperception and productionperception and productionperception and production

Carolina Serra & Sónia Frota

UFRJ/CAPES (Brazil),Universidade de Lisboa (Portugal)( g )

IntroductionIntroduction & Background& Background

It has long been known that the location of intonationalboundaries is somehow related to the location of syntacticyboundaries.Prosodic hierarchy theory captures this relation:

t h l i i i l + h l– syntax-phonology mapping principles + phonology properconstraints >> predict preference patterns of intonationalphrasing in production and comprehension (Selkirk, 2000;Watson & Gibson 2005; for Romance languages Elordieta etWatson & Gibson, 2005; for Romance languages, Elordieta etal., 2005; Frota et al., 2007).

Previous descriptions of intonational breaks within Romance(b d d h i l d )(based on read speech isolated sentence corpora):– the main boundary cues are a H boundary tone and a F0 rise

either on the stressed or on the boundary syllable (Frota et al.,y y (2007).

The The PhPPhP and the IP in Portugueseand the IP in Portuguese

Phonological Phrase (PhP) construction:– Lexical head + all elements to the left within the Maximal

projection of Lex + following XP complement containing just oneprojection of Lex + following XP complement containing just onePW (Frota 2000, Tenani 2002).

– In BP, unlike EP, the PhP is characterized by the regularoccurrence of a pitch accent in the most prominent element(Frota & Vigário, 2000; Tenani, 2002).

Intonational Phrase (IP) construction:Intonational Phrase (IP) construction:– Strings not structurally attached to the sentence tree / any

remaining sequence of adjacent PhPs in a root sentence are( ) S ( ) Cmapped onto IPs (Frota 2000). Akin to Selkirk’s (2005) Comma

Phrase.– In BP & EP: a nuclear contour and a potential pause ; In EP: pre-& a uc ea co ou a d a po e a pause ; p e

boundary lengthening (Frota 2000, 2003; Tenani, 2002; Vigário2003).

Present ResearchPresent Research

GoalsGoals(1) To capture the relation between prosodic breaks as( ) p p

established by prosodic hierarchy theory and perceivedprosodic boundaries in read and spontaneous speech;

(2) To test the effects of syntactic boundary rank and of(2) To test the effects of syntactic boundary rank and ofprosodic distance between boundaries on boundaryperception;p p

(3) To describe the phonetics and phonology ofintonational breaks, namely the inventory of tones andthe way they are realized;the way they are realized;

(4) To assess the most prominent cues to the perception ofprosodic breaks in read speech and spontaneousspeech.

Main contributions to the fieldMain contributions to the field

* New data: Brazilian Portuguese (BP)

* Contribution to the cross-linguistic knowledge about theplacement and shape of prosodic boundaries

* Contribution to the understanding of the way prosodicboundaries are realized and perceivedboundaries are realized and perceived

* Contribution to the knowledge about prosodicdiff ( di h i d h i fdifferences (prosodic phrasing and the various types ofboundary cues) across speech styles (e.g. Blaauw 1994;Jurafsky 2008)Jurafsky 2008)

MethodMethod

CorpusCorpus:: 5 spontaneous speech (SS) and 5 read speech(RS) passages of about 2 minutes each(RS) passages of about 2 minutes each.

SpeakersSpeakers:: (1) Production - 5 female speakers.(2) Perception - 11 listeners.( ) p

University students, born in Rio de Janeiro, aged 22-38.Procedure:Procedure:(1) Production – Interviews (SS); Reading of a passage taken from the interview (RS). Passages fully annotated for phonological and intonational phrasing (for Portuguese Frotaphonological and intonational phrasing (for Portuguese, Frota2000, Tenani 2002, Fernandes 2007), to define the placement of predicted prosodic breaks.

MethodMethod

ProcedureProcedure::

(2) Perception• Listeners had to signal the prosodic breaks they

perceived in each passageperceived in each passage. • Breaks should be signaled on a piece of paper

containing the orthographic transcription of each fpassage, without any kind of punctuation marks.

• To assess consistency, listeners repeated the task in 2 sessions (for each session the items were randomizedsessions (for each session the items were randomized differently).

• The task was preceded by a training phase. Instructions ti d th t ki d f b k h ld b k dmentioned that any kind of break should be marked.

Measurements: Measurements:

● Location of silent pauses● Lengthening patterns in the nuclear word: duration oft d d t t d ll bl l ti t th t dstressed and post-stressed syllables relative to the pre-stressed

syllable, expressed as percent values● F0 variation in the nuclear word: F0 measurements obtainedat the peak of maximum intensity in the syllable (PMI) were thebasis for the computation of F0 variation between stressed andpre-stressed syllables, and F0 variation between post-stressedp y , pand stressed syllables, both expressed as percent values● Type & frequency of nuclear contours (nuclear PA andboundary tone)boundary tone)● Type & frequency of syntactic boundary rank of (not)perceived boundaries● Distance between (not) perceived boundaries in nº ofsyllables and prosodic words (PWs)

C iC i di t d h i ddi t d h i dComparison: Comparison: predicted phrasing and predicted phrasing and perceived phrasingperceived phrasing

● Perceived breaks = Breaks perceived by at least 8of the 11 listeners

● Properties of predicted AND perceived breaks

● Properties of predicted but NOT perceived breaks

● Properties of perceived breaks NOT predicted as IPs

R lt d Di iR lt d Di iResults and DiscussionResults and Discussion

More predicted IPs are perceived as breaks in RSth i SS

Perceived prosodic breaksoccur at IP-boundaries in

than in SSboth styles

Distribution of PhP-edges perceived and not perceived

PP-edges perceived

PP-edges not perc Totalperceived not perc. Total

RS 2 (0,5%) 436 (99,5%) 438

SS 10 (2,3%) 428 (97,7%) 438

Total 12 (1,4%) 864 (98,6%) 876

R lt d Di iR lt d Di iPhoneticPhonetic andand PhonologicalPhonological PropertiesProperties ofof predictedpredicted &&

Results and DiscussionResults and Discussion

perceivedperceived // nonnon--perceivedperceived breaksbreaks• The main cue to the perception of a break is thepresence of a pause across speakers and for bothpresence of a pause, across speakers and for bothstyles.

R lt d Di iR lt d Di iPhoneticPhonetic andand PhonologicalPhonological PropertiesProperties ofof predictedpredicted &&

Results and DiscussionResults and Discussion

perceivedperceived // nonnon--perceivedperceived breaksbreaks• Pauses are longer next to perceived IP-edges,across speakers and for both styles Perceived PhPacross speakers and for both styles. Perceived PhP-edges behave as IP-edges (> they were realized as IPs)

R lt d Di iR lt d Di iPhoneticPhonetic andand PhonologicalPhonological PropertiesProperties ofof predictedpredicted &&

Results and DiscussionResults and Discussion

perceivedperceived // nonnon--perceivedperceived breaksbreaks

• Lengthening patterns and F0 properties at the IP-edge:Speaker-dependent strategies

• The bulk of lengthening goes on the stressedsyllable (as previously reported for EP in Frota 2000). Itsyllable (as previously reported for EP in Frota 2000). Itmay be a cue to perception, dependent on the speaker(e.g. Sp1, SS). Overall, bigger lengthening contrasts inSS P i d PhP d i SS b h IP dSS. Perceived PhP-edges in SS behave as IP-edges(> they were realized as IPs)

Phonetic and Phonological Properties of predicted & Phonetic and Phonological Properties of predicted & perceived / nonperceived / non perceived breaks:perceived breaks: lengtheninglengtheningperceived / nonperceived / non--perceived breaks: perceived breaks: lengtheninglengthening

RS

SS

* only two cases

SS

Phonetic and Phonological Properties of predicted & Phonetic and Phonological Properties of predicted & perceived / nonperceived / non perceived breaks:perceived breaks: F0 propertiesF0 properties

• F0 variation may be a cue to perception, but also

perceived / nonperceived / non--perceived breaks: perceived breaks: F0 propertiesF0 properties

dependent on the speaker (e.g. Sp2, RS). Speaker’sstrategies seem to either favour lengthening or F0 variation(e g Sp1 vs Sp2) Overall larger F0 variation (on the(e.g. Sp1 vs. Sp2). Overall, larger F0 variation (on thestressed syllable) in RS.

RS

* only two casesSS

Phonetic and Phonological Properties of predicted & Phonetic and Phonological Properties of predicted & perceived / nonperceived / non perceived breaks:perceived breaks: Nuclear contoursNuclear contours

• H+L* L% is the most common contour in both styles, but

perceived / nonperceived / non--perceived breaks: perceived breaks: Nuclear contoursNuclear contours

with different frequencies (RS > SS)

IP-edge with perceived break, Sp 3 – RS [a gente podia ter ficado mais um

RS

pouquinho] IPwe could have stayed longer

Phonetic and Phonological Properties of predicted & Phonetic and Phonological Properties of predicted & perceived / nonperceived / non perceived breaks:perceived breaks: Nuclear contoursNuclear contours

• Falling/low nuclei & Low boundary are dominant in RS

perceived / nonperceived / non--perceived breaks: perceived breaks: Nuclear contoursNuclear contours

(against previous results for read isolated sentences – Frota et al. 2007);in SS the distribution of falling/low and rising/high nuclei, aswell as low/high boundaries is about evenwell as low/high boundaries is about even.

IP-edge with perceived b k S 1 RS / SS

RS

break, Sp 1 – RS / SS [foi um período muito ruim pra mim] IP It was a very bad time |It was a very bad time | for me |SS

Phonetic and Phonological Properties of predicted & Phonetic and Phonological Properties of predicted & perceived / nonperceived / non perceived breaks:perceived breaks:perceived / nonperceived / non--perceived breaks: perceived breaks: Nuclear contoursNuclear contours Boundary tonesBoundary tones

RS RS

SS SS

R lt d Di iR lt d Di iPhoneticPhonetic andand PhonologicalPhonological PropertiesProperties ofof predictedpredicted &&

Results and DiscussionResults and Discussion

perceivedperceived // nonnon--perceivedperceived breaksbreaks :: IPIP sizesize• Perceived IPs are longer than non perceived IPs, acrossspeakers and styles (both in nº syll and nº PWs) >>speakers and styles (both in nº syll and nº PWs) >>

Size affects i

RSperception**

*

SSNº syll already shown to be relevant to IP formation as a

SS

size condition: >8 (>60%); >12 (>90%) Elordieta et al 2005

R lt d Di iR lt d Di iPhoneticPhonetic andand PhonologicalPhonological PropertiesProperties ofof predictedpredicted &&

Results and DiscussionResults and Discussion

perceivedperceived // nonnon--perceivedperceived breaksbreaks :: IPIP sizesize• nº PWs

Size affects i

RSperception**

**(*)

SSSS*

R lt d Di iR lt d Di iDistance affects

perception

PhoneticPhonetic andand PhonologicalPhonological PropertiesProperties ofof predictedpredicted &&

Results and DiscussionResults and Discussion perception

perceivedperceived // nonnon--perceivedperceived breaksbreaks :: distancedistance• Distance between Perceived breaks is longer thanbetween perceived and non perceived breaks acrossbetween perceived and non-perceived breaks, acrossspeakers and styles (both in nº syll and nº PWs)

RS RS

* *

*(*)

(*) ***

**

SS SSSS SS***

*** ** ***

***

* *** ** ***

R lt d Di iR lt d Di iSyntacticSyntactic boundaryboundary rankrank

Results and DiscussionResults and Discussion

• Perceived IP-edges are mostly matrix sentences edges,across styles >> Par, DE, XP >> Top, IS >> C

- S boundary is more robustly perceived as a major prosodicboundary (e.g. Collier & ‘t Hart 1975, Cole 2008)- But no neat gradual effect of syntactic rank suggesting that- But no neat gradual effect of syntactic rank, suggesting thatmapping predictability (Edge S>Edge IP) has a role to play(phrasing of Par, DE, Top is affected by length effects/prosodic subordination)

SummarySummary

P i d b k t IP d dl f t l• Perceived breaks occur at IP-edges, regardless of style• More predicted IPs NOT perceived in SS (63% vs. 36%)• Properties of predicted & perceived IPs / not perceived:p p p p1. a pause is the strongest cue, across speech styles2. lengthening (stressed syll.) is speaker-dependent and

characterises especially SScharacterises especially SS3. F0 variation (stressed syll.) is speaker-dependent and

characterises especially RSp y4. Nuclear contours: more falling contours in RS; more

contour variation (token, not type) in SS5 Longer IPs are more robustly perceived across styles5. Longer IPs are more robustly perceived, across styles6. Distance between IPs promotes the perception of a

break, across speech styles7. Syntactic rank shows a main effect of S (mapping

predictability), across speech styles

ConclusionConclusion• SS shows more variability in the relation between predictedSS shows more variability in the relation between predicted,perceived and produced phrasing (as expected). The variabilityemerges from a more even distribution of the same types of cues,that seem to play similar roles in both styles This eventhat seem to play similar roles in both styles. This evendistribution may be an important factor behind the lowerperception of predicted breaks in SS.

U lik i i t ( H ll & K di H ifi 1991• Unlike in previous reports (e.g. Howell & Kadi-Hanifi 1991,Blauuw 1994), BP data does not show marked differences in typebetween styles (such as pauses mainly in SS or at differentlocations, or a tendency to produce much more phrases in one ofthe styles relative to the other).• In BP, predicted IP-phrasing is fairly robust across styles: 87%p p g y yall predicted IPs were realized as such (94% in RS; 81% in SS);only 0.8% of predicted PhPs were perceived as breaks anduttered as IPs (0.3% in RS; 1.3 in SS).uttered as IPs (0.3% in RS; 1.3 in SS).

THANK YOU!THANK YOU!

ReferencesReferencesReferencesReferencesBLAAUW, E., 1994. The contribution of prosodic boundary markers to the

l diff b d d h S hperceptual difference between read and spontaneous speech. SpeechCommunication 14, p. 359-375.COLE, J. et al., 2008. The role of syntactic structure in guiding prosody perceptionin spontaneous speech. Experimental and Theoretical Advances in Prosody,in spontaneous speech. Experimental and Theoretical Advances in Prosody,Cornell University, April 11-13.COLLIER, R. & t´HART, J., 1975. Integrating different levels of intonation analysis.In: Journal of Phonetics, 3, p. 235-255.ELORDIETA G t l 2005 S bj t bj t d i t ti l h i iELORDIETA, G. et al., 2005. Subjects, objects and intonational phrasing inSpanish and Portuguese. Studia Linguistica. 59 (2-3), 110-143.FERNANDES, F. R., 2007. Ordem, Focalização e Preenchimento em Português:Sintaxe e Prosódia. Tese de Doutorado em Lingüística. Campinas: LEL/UNICAMP.Sintaxe e Prosódia. Tese de Doutorado em Lingüística. Campinas: LEL/UNICAMP.FROTA, S., 2000. Prosody and focus in European Portuguese. Phonologicalphrasing and intonation. New York: Garland Publishing.________., 2003. The phonological status of initial peaks in European Portuguese.C t l J l f Li i ti 2 133 152Catalan Journal of Linguistics, n. 2, p. 133-152.FROTA, S. & VIGÁRIO, M., 2000. Aspectos de prosódia comparada: ritmo eentoação no PE e no PB. In: CASTRO, R. V. & BARBOSA, P. (eds.). Actas do XVEncontro Nacional da Associação Portuguesa de Lingüística, v. 1. Coimbra: APL,Encontro Nacional da Associação Portuguesa de Lingüística, v. 1. Coimbra: APL,p. 533-555.

ReferencesReferencesFROTA, S. et al., 2007. The phonetics and phonology of intonational phrasing inRomance. In: Pilar Prieto, Joan Mascaró & Maria-Josep Solé (eds). Prosodic andSegmental Issues in (Romance)Phonology. John Benjamins (Current Issues inLinguistic Theory).HOWEL, P. & KADI-HANIFI, K., 1991. Comparison of prosodic properties betweenread and spontaneuos speech. Speech Communication 10, p. 163-169.JURAFSKY D 2008 Predicting and detecting prosodic prominenceJURAFSKY, D., 2008. Predicting and detecting prosodic prominence.Experimental and Theoretical Advances in Prosody, Cornell University, April 11-13.SELKIRK, E., 2000. The interaction of constraints on prosodic phrasing. In:Prodody: Theory and Experiment, ed. Merle Horne, 231-262. Dordrecht: KluwerAcademic Publishing.___________. 2005. Comments on Intonational Phrasing in English. In Prosodies:Selected papers from the Phonetics and Phonology in Iberia Conference,2003 (Phonetics and Phonology Series ) ed S Frota M Vigario M and J2003.(Phonetics and Phonology Series.) ed. S. Frota, M. Vigario, M. and J.Freitas, Mouton de Gruyter.TENANI, L.E., 2002. Domínios prosódicos do Português do Brasil: implicaçõespara a prosódia e para a aplicação de processos fonológicos. Tese de Doutoradoem Lingüística. Campinas: LEL/UNICAMP.VIGÁRIO, M., 2003. The Prosodic Word in European Portuguese. Berlin/ NewYork: Mouton de Gruyter.WATSON D & GIBSON E 2005 Intonational phrasing and constituency inWATSON, D. & GIBSON, E., 2005. Intonational phrasing and constituency inlanguage production and comprehension. Studia Linguistica. 59 (2-3), 279-300.