<<

View metadata, citation and similar papers at core.ac.uk brought to you by CORE

provided by Repository of the Academy's Library

Prosodic (non-)realisation of broad, narrow and contrastive in Hungarian: a production and a perception study

Katalin Mady´

Research Institute for Linguistics, Hungarian Academy of Sciences, Budapest, Hungary [email protected]

Abstract intermediate status such as Neapolitan Italian. In Spanish that is prosodically non-plastic, prosodic focus marking is optional, In languages with variable focus positions, prominent ele- whereas it is obligatory in Neapolitan Italian, both in the default ments tend to be emphasised by prosodic cues (e.g. English). If (sentence-final) and the non-default focus position. a language prefers a given prosodic pattern, i.e. sentence-final nuclear accents, like Spanish, the prosodic realisation of broad At the same time, prosodic focus marking in prosodically focus might not differ from that of narrow and contrastive fo- plastic languages is not necessarily consistent. According to a cus. The relevance of prosodic focus marking was tested in study by Baumann et al. [5] in which they studied focus reali- Hungarian were focus typically appears in front of the finite sations in German, speakers applied different strategies to mark verb. Prosodic cues such as f0 maximum, f0 peak alignment, the degree of prominence: they used a combination of para- segment duration and post-verbal deaccentuation were tested in metric means such as lengthening, f0 range and maximum and an experiment with read question and answer sequences. While its position and of categorical means such as different pitch ac- narrow and contrastive focus triggered post-verbal deaccentu- cents. In their study, only one out of 6 speakers made use of all ation, none of the gradual measures distinguished focus types available strategies. consistently from each other. A subsequent perception experi- Similarly to Spanish, Hungarian is a prosodically non- ment was conducted in which the same sentences without post- plastic language with syntactical focus marking, and it is verbal units were to be judged for their naturalness. F0 maxi- prosodically left-headed [4]. Word order reflects a logical struc- mum, f0 peak alignment and accent duration were manipulated. ture: according to [6], Hungarian sentences can be divided into Naturalness scores revealed a sequence narrow > contrastive > a topic and a part. The topic position includes units broad focus, i.e. a preference for narrow focus contexts com- about which a prediction is made in the predicate part of the pared to contrastive and broad focus ones, while the manipu- sentence. E.´ Kiss claims that the first major constituent of the lated prosodic parameters had no effect on the scores. It is con- predicate has to be accented, and it bears the heaviest accent cluded that prosodic focus marking in Hungarian is optional and in the sentence. According to her view, the nuclear accent is pragmatic rather than grammatical and syntax-related. located at the left edge in Hungarian, but its domain is the pred- Index Terms: prosody, focus, information structure, Hungarian icate and not the entire sentence. As a consequence, the topic has to be treated as a separate intonational phrase (IP) as was 1. Introduction suggested by [7, 8, 9]. However, this view is problematic since an IP boundary is not necessarily present in each sentence which Information that is relevant for communication is emphasised contains an accented topic and a predicate. in some way in most languages. Prominence can be marked by prosody, syntax, , or by their combination. The In Hungarian, foci are located immediately before the finite flexibility of word order and the variability of prosodic patterns verb. If the sentence has broad focus, word order is relatively is interrelated: languages with a relatively rigid word order like free, also with respect to the pre-verbal position. It is assumed English tend to use prosodic prominence marking, whereas lan- that focus is mainly expressed by the obligatory deaccentuation guages with a preference for a certain prosodic pattern such as of the finite verb and of post-verbal elements [10]. Spanish [1] or Czech [2] move the focussed element to the po- The goal of the present paper is to investigate the role of sition that bears strong prominence by default. While the de- prosodic cues in focus marking in Hungarian. Given the strict fault nuclear accent position is sentence-final both in English syntactic marking of foci, it is expected that prosodic cues are and most Romance languages, the usual way to express em- not necessary to express focus prominence. First, a production phasis in English would be to shift the nuclear accent if the experiment is presented (see [11] for an outline in Hungarian) prominent element is non-final, whereas in Spanish the promi- which serves as a basis for a perception study. The research nent unit would be shifted to the sentence-final position. Lan- questions are the following: (1) is focus marking supported by guages that prefer prosodic prominence marking are referred prosodic cues in Hungarian? (2) Are there differences in the to as prosodically ‘plastic’, since they show variable prosodic distribution of pitch accents? (3) Do certain pitch accent types patterns, whereas languages that tend to maintain a certain contribute to the distinction? The study includes the investiga- prosodic pattern but allow for syntactic flexibility are referred tion of the focus, the topic, the verb, and the post-verbal part of to as prosodically ‘non-plastic’ languages [3, 4]. the comment. Three focus types are compared: sentences with Prosodic plasticity is gradual with respect to the rigidity of broad focus are all-new sentences, i.e. not a single constituent, the maintenance of word order vs. the prosodic pattern. For ex- but the entire sentence is focussed. The term narrow focus is ample, [1] showed that word order shift is obligatory in Spanish used for -new constituents that answer a wh-question, or Catalan, whereas the tendency is weaker in languages with an while contrastive focus expresses explicit . 2. Production experiment Categorical cues were analysed by means of χ2 tests. For parametric cues, generalised linear mixed models (lme4 pack- Two target sentences were created with a so-called verbal mod- age in R) were used with focus type as a fixed effect and speaker, ifier (VM). This unit appears immediately before the finite verb sentence and repetition as random effects. Significance level even if it is not focussed, thus it is possible to use string- was set to α = 0.05. identical sentences with broad, narrow and contrastive focus de- pending on the context. 2.1. Results Sentence 1: A lanyom´ Nemetorsz´ agba´ ment munkat´ keresni. 2.1.1. Categorical cues the daughter-my Germany-to went job-acc search-to Topics were deaccented in 71% of all cases, and if a pitch ac- ‘My daughter went to Germany to look for a job.’ cent was present, it was usually rising (71%). According to a Sentence 2: pairwise , the distribution of pitch accent categories Marianna Malt´ an´ nyaral Monival.´ showed no effect of focus type (p > 0.2). Foci were accented in Marianna Malta-on has-holiday Moni-with´ all but 2 cases and carried mostly falling accents (H+L*: 82%, ‘Marianna is on holiday in Malta with Moni.’´ H*+L: 8%, L*: 6%). Their distribution did not differ signifi- cantly across focus types (p > 0.1). The verb was deaccented Since lexical stress is word-initial in Hungarian, the first in all but 6 cases (93%). post-verbal units were deaccented in syllable of each content word was the potential domain for ac- about 50% of the cases in the narrow and contrastive focus con- centuation. Sentences were displayed to participants on a screen dition and in 32% in the broad focus condition. If there was an using SpeechRecorder [12] preceded by a question that elicits accent, it usually had an H+L* pattern. broad, narrow or contrastive focus. Speakers were asked to read These results show a uniform prosodic pattern for all sen- the target sentence as an answer to the following questions: tences. Pitch accents did not differ according to focus type. The 1. broad focus: (1): How is it going? How is Erika? (2): only difference was that the deaccentuation of post-verbal el- How are you ()? And how is your colleague? ements was more frequent after narrow and contrastive focus. 2. narrow focus: (1): So where is Erika looking for a job? This shows that [10]’s claim about the obligatory accent dele- (2): So where did your colleague and her friend decide tion after a focus can be observed as a tendency, but not as a to go for holiday? strict rule. Erika is looking for a job in Pest, 3. contrastive focus: (1): 2.1.2. Parametric cues right? (2): Your colleague and her friend went to Sweden in the end, didn’t they? Separate analyses were run for the tonal cues of the topic, focus and the post-focal units. Only comparisons with a significant 7 students (6 females, 1 male, mean age 21 years) partici- difference are reported here. pated in the experiment that was carried out in a sound-treated The f0 minimum of the topic was aligned earlier and the room at the Department of Phonetics, RIL HAS. Each speaker maximum aligned later before contrastive focus. These CVC read each target sentence in three focus conditions with 6 repe- sequences were also significantly longer than those with broad titions in randomised order, totalling in a stimulus set with 252 and narrow focus, thus the difference in tonal alignment might items. The contexts were merged with the stimuli for another be a result of lengthening. The f0 peak on the focus was aligned experiment serving as distractors here. later if the focus was contrastive, but no lengthening of the ac- The following parameters were analysed in the pre-verbal cented CVC sequence was observed. Fig. 1 shows the f0 max- part of the sentence, i.e. on the topic and the focus: imum alignment for each speaker where the percentage value Categorical cues: specifies the point at which f0 starts to fall within the focus- • occurrence of deaccented content words, initial CVC (a higher percentage means a later maximum align- ment). Interestingly, the alignment is later in contrastive focus • distribution of pitch accent types. than in broad focus, while the marking of narrow focus is highly Parametric cues within word-initial CVC sequences (irrespec- speaker-dependent: alignment is earlier than in broad focus in tive of the syllable structure): some speakers’ production and later than in contrastive focus in other speakers’ data. • f0 maximum, The results show that broad, narrow, and contrastive focus • f0 minimum, were not distinguished consistently. Broad focus differed from • f0 range, narrow and contrastive focus by the more frequent accentuation of the post-verbal constituent. Neither pitch accent categories • f0 slope, nor the occurrence of deaccentuation of the topic revealed a sig- • duration of the f0 rise or fall, nificant difference between focus types. There were two param- • alignment of the target within the CVC sequence, eters that distinguished contrastive focus from the other two: they were preceded by longer topic CVC duration and a less • duration of the CVC sequence. steeper f0 rise on the accented syllable, and the slope for the The analysis of the post-focal part was restricted to the cat- fall on the focus started later leading to a longer plateau between egorical cues listed above. the topic and the focus. Again, there was a large inter-speaker Pitch accents were labelled by two labellers using tonal la- variation. bels such as L*, H*, and their combinations. F0 was measured In the second experiment it was investigated whether tonal in semitones with a value of 100 Hz. The duration cues can enhance focus identification. Tonal alignment was of the accented syllable was normalised to the entire sentence tested along with two parameters that had been found to trig- length. ger the perception of emphasis in several languages. / ra’ SL ytei module: synthesis using PSOLA manipulated Praat’s were parameters following The realisations. (/ accented different the were sentences Since two maxi- the manipulation. and in for sequences minimum CVC used measured were the values and mum speaker, model the of ob- was as focus, the data. on production accent the falling in a served rising and a topic had the sentences on Both accent manipulation. for basis the as lected duration the and of maximum, sequence vowel. f0 CVC accented the the accented) of of (= localisation the initial following: VM, the the the in focus were maximum contrastive manipulation f0 and for the narrow parameters broad, The the setting. in times five speaker were questions preceding The Malta’. in accordingly. adjusted holiday on is ianna and (de)accentuation. Germany’, to their went of daughter were influence sentences post-verbal an avoid any Thus, involve to order not in did units, produc- they the because in experiment those from focus. tion slightly contrastive differed or sentences narrow target broad, The as interpreted mod- verbal be the could which were in ifier pairs sentences answer and target question experiment, as presented production the to Similarly methods and Material 3.1. 2. sentence bottom: 1, 100%: sentence duration, Top: CVC duration. of CVC beginning of end 0%: focus. the of sequence 1: Figure ma:l h aaeest emnpltdwr nlsdi h data the in analysed were manipulated be to parameters The se- was focus narrow with sentence each of realisation One female a with recorded were pairs answer and Question ) h devle eeaatdt h sentence-specific the to adapted were values edge the /), oiino h 0mxmmwti h cetdCVC accented the within maximum f0 the of Position

position of f0 maximum (%) position of f0 maximum (%) .Preto experiment Perception 3. 5 10 15 20 25 30 35 0 5 10 15 20 25 30 f0 maximum alignmentinCVCsequence f0 maximum alignmentinCVCsequence broad broad focus type, sentence2,linesrepresentspeakers focus type, sentence1,linesrepresentspeakers l A no N anyom ´ narrow narrow ainaM Marianna emetorsz ´ contrastive contrastive alt ´ gament agba ´ nnyaral an ´ ne:m ‘Mar- ‘My vs. / goe nfrhraayi.Teefc ffcstp a highly was type focus thus of ( was effect significant and The effect analysis. random further a in ignored as significant not was Repetition Results 3.2. α the the by in tested random vided as were repetition Interactions and duration sentence vowel effects. speaker, and and alignment effects fo- slope fixed variable, f0 as dependent maximum, f0 the type, as cus scores naturalness with models years). males, 31 (18 age experiment mean the females, in 10 stimulus participated the answer listeners to and 28 added was question material. distractors 96 of of amount same set The a pairs. in totalling participant, each to Each sentences. ments across varied questions that (3 combination strength at- accent pay or to to synthesis the instructed tention of were quality participants the unnaturalness, as influence semantic such the exclude factors to a other order using potential In question, of 7. the and to 1 between sounded scale answer asked discrete the were natural they how and pair, rate Par- answer to and focus. question contrastive a or heard narrow ticipants broad, ques- either a elicited by preceded that was tion follow- sentence manipulated the each to according design: ing presented were Stimuli sentence. each rgee infiatyhge auans crs( scores maximum f0 naturalness higher higher a significantly with targets triggered condition, focus ( narrow level significance the condition, reached effects focus fixed broad the the of In none type. focus each within vestigated is answers experiment. focus follow-up a narrow in of tested preference being the communica- currently for spontaneous reason in The rare tion. Again, relatively natural. is less contrast sound might explicit question colleague?” all-new your everyday is an “How in to like seeking reply same information the of whereas way communication, natural a is as holiday?” such one: a information to pragmatic discourse-new answer a contains the is that explanation other be- utterance choice The an the types. influenced other crucially focus parameters that tween were ones there analysed that the possible than is It a reply. manipu- focus to underwent row answers that both utterances were original lation the first, each for reasons: Means 5). 2. Fig. in median: shown are 4.82, participant 4.17, interme- (mean: received (mean: answers judgements natural focus least diate score: contrastive as (mean while focus 4), natural broad median: most and 6), as focus median: narrow 5.2, elicited that tions 0 = nlsswspromduiggnrlsdlna mixed linear generalised using performed was Analysis of set a Thus, usqety h feto h oa aiuainwsin- was manipulation tonal the of effect the Subsequently, two have can context focus narrow the for preference The h uaino h cetdvwl(etne1 5and 85 1: (sentence vowel accented the of duration the • the where point time the i.e. peak, f0 the of alignment the • (sen- syllable accented the on maximum f0 of value the • . × 2 s etne2 4 n 8 ms). 185 and 145 2: sentence ms, 125 sentences), both and in before vowel ms accented the (20 of started onset accent the H+L* after the in fall pitch sentences Hz), 150 all 2: sentence through Hz), Hz, 260 constant 185 and 1: 240 being (sentence 2: minimum sentence f0 Hz, 270 the and 250 1: tence 05 . oe durations vowel 2 < p car wh 0 2 akg nR infiac ee a e to set was level Significance R. in package . qeto Weeddyu olau oon go colleague your did “Where -question 001 × 2 .Priiat ugdasest ques- to answers judged Participants ). × × 8 = 2 × 0maxima f0 2 wh etne)wspae twice played was sentences) 2 obntoswscetdfor created was combinations qeto hteiie nar- a elicited that -question Anova × 0soealign- slope f0 2 ucinpro- function > p p 0 = 0 . 2 . .In ). 02 ). h iens ftebcgon nomto eeinvestigated. were information background with the along of types focus the study, this In conclusions. contradicting pro- the focus, in contrastive present not for experiment. higher was duction not to tendency lead corresponding but f0 a those narrow, higher whereas than time, for scores same scores the naturalness At higher naturalness get one. earlier not an did with fall with f0 samples later interpretation: a contrastive enhance not does ment production. speakers’ less many is in focus focus narrow broad whereas than alignment, marked tonal by fo- assumption. marked contrastive is this only results cus variation, support inter-speaker the strong not However, a do from Apart experiment contrast. production express the since not greater, of if be be- does will that difference types focus two the expected other broad then the is marking, and It focus focus broad in tween role as- Hungarian. a are in plays both exhaustive prosody and be contrast, to implicit sumed carries focus contrastive, narrow explicitly contrastive is while and focus narrow Contrastive emphasis. both bear structure, focus information of terms In interaction significant. an duration was by vowel effect and signalised high maximum long, was f0 was maximum between duration f0 vowel if a the scores was and naturalness there of condition, increase focus contrastive slight tonal the of In independently almost manipulations. replies natural apart focus relatively scores, was narrow as 0.9 words, there judged than other were If larger In not (1.5). 3). was participant Fig. one it from preference, (see the other in each bias from a differ not did pants ( ecswt ihf eevdhge crsfo 7listeners 17 from scores higher received sen- f0 ( participants: high among with variation tences large a was there However, participant. medians. each overall the for to conditions refers focus contrastive and 2: Figure mean 0 = naturalness scores (mean) eetsuyb 1]wihi ne ulcto ed to leads publication under is which [13] by study recent A align- tonal later that reveals experiment perception The . 46 1 2 3 4 5 6 7 = , en fntrlessoe o h ra,narrow broad, the for scores naturalness of Means sd broad MED − 0 0 = . 37 lines representmeansfor eachlistener , . sd 43 Scores for focus type .Discussion 4. n oe crsfo participants 8 from scores lower and ) 0 = . 31 narrow MED ,wiersosso partici- 3 of responses while ), p 0 = . 09 contr MED oother No . MED A)frtehl ihteceto ftesiuu e o the for HAS. set RIL at stimulus Phonetics the of of Department the creation and the study with production thank help to the like for HAS) would I the by Bal Foundation. supported Humboldt was stay von conference Alexander The Research Scientific 101050). Hungarian (PD the Fund by supported was work This in position pre-verbal the Hungarian. in grammatical reflect structure to information seem was or not other categories do the foci on on for focus cues focus contrastive prosodic and broad found, no narrow between and Since hand distinction one systematically. the consistent utilised a not Hun- for is in evidence production marking and focus the optional prosodic in is that present concluded garian not is it natural- was Thus higher foci to data. narrow lead for that scores At maximum f0 ness experiment. higher perception time, the same in the replicated be this tonally, not marked could are parameters effect foci investigated contrastive that (many) per- showed the focus and the of production of one in only results while There the ception: Hungarian. between in asymmetry marking an focus was for cues tonal of relevance focus. and broad narrow before before than frequent focus where contrastive more data our not with was line following deaccentuation in the topic not of is tendency prominence this stronger However, a focus. this trigger deac- of that might deaccentuation as assumed the topic categorised Thus, be the be guidelines. to our rather is to would according it cented realisation [13] topic of of paper of kind the demonstration con- the in the on topics signalled based However, rising rising topic. contrastive that the on found of study trastivity [14] a Hungarian, in in because glance, topics first seems the This at foci. contrastive surprising ac- found by followed pitch was when rising found semitone with were topics 1 cents more about Besides, f0 syllables. focussed of in increase significant a Here, participant. each medians. overall for the condition to focus refers narrow the in ima 3: Figure z Sur azs ´ naturalness scores (mean) h eut fbt xeiet hwwa vdnefrthe for evidence weak show experiments both of results The 1 2 3 4 5 6 7 en fntrlessoe o ihadlwf max- f0 low and high for scores naturalness of Means ni(oetmpoetgatn.21-1,RIL 2011-013, nr. grant project (Momentum anyi ´ MED low .Acknowledgements 5. lines representmeansfor eachlistener Scores for narrow focus f0 maximum MED high MED 6. [1] T. L. Face and M. D’Imperio, “Reconsidering a focal typology: evidence from Spanish and Italian,” Italian Journal of Linguistics, vol. 17, no. 2, pp. 271–289, 2005. [2] T. Dubedaˇ and K. Mady,´ “Nucleus position within the intonation phrase: a typological study of English, Czech and Hungarian,” in Proc. Interspeech, Makuhari, Japan, 2010, pp. 126–129. [3] E. Vallduv´ı, “The role of plasticity in the association of focus and prominence,” in Proc. ESCOL 7, 1991, pp. 295–306. [4] D. R. Ladd, Intonational phonology, 2nd ed. Cambridge: Cam- bridge University Press, 2008. [5] S. Baumann, M. Grice, and S. Steindamm, “Prosodic marking of focus domains – categorical or gradient?” in Proc. 3rd Speech Prosody, Dresden, Germany, 2006, pp. 301–304. [6] K. E.´ Kiss, The syntax of Hungarian. Cambridge: Cambridge University Press, 2002. [7] I. Vogel and I. Kenesei, “The interface betwen phonology and other components of grammar: the case of Hungarian,” Phonol- ogy Yearbook, vol. 4, pp. 243–263, 1987. [8] L. Varga, Intonation and stress: evidence from Hungarian. Bas- ingstoke & New York: Palgrave Macmillan, 2002. [9] L. Hunyadi, Hungarian sentence prosody and universal grammar: on the phonology–syntax interface. Frankfurt/Main: Lang, 2002. [10] P. Siptar´ and M. Torkenczy,¨ The Phonology of Hungarian. Ox- ford: University Press, 2000. [11] K. Mady,´ “A fokusz´ prozodiai´ jelol¨ ese´ felolvasasban´ es´ spontan´ beszedben´ [Prosodic marking of focus in read and spontaneous speech],” in Beszed,´ adatbazis,´ kutatasok´ , M. Gosy,´ Ed. Bu- dapest: Akademiai´ Kiado,´ 2012, pp. 91–107. [12] C. Draxler and K. Jansch,¨ “SpeechRecorder – a universal platform independent multi-channel audio recording software,” in Proc. In- ternational Conference on Language Resources and Evaluation, Lisbon, 2004, pp. 559–562. [13] S. Genzel, S. Ishihara, and B. Suranyi,´ “The prosodic expression of focus, contrast and givenness: A pro- duction study of Hungarian,” Lingua, in press, http://dx.doi.org/10.1016/j.lingua.2014.07.010. [14] B. Gyuris and K. Mady,´ “Contrastive topics between syntax and in Hungarian: an experimental analysis,” Papers of Chicago Linguistic Society, p. 15 p., 2014.