Structural analysis of prosodic patterns : the case of excessive prolongations in Israeli

This paper proposes a new framework for prosodic pattern analysis, based on the study of excessive prolongation in spontaneous Israeli Hebrew. In order to reveal whether it is a random phenomenon or a predictable prosodic pattern, a multi-layer linguistic analysis was conducted. First, the phenomenon was taken out of its canonical research framework as a type of speech disfluency. Second, its acoustic characteristics were defined, and the phonological environments of these prolongations were accounted for. Finally, prolongations and their interface with the syntagmatic layer were analyzed. The proposed framework can serve as a format for other prosodic patterns as well.


Introduction
An effective approach to the study of prosody in spoken language seeks to identify prosodic patterns and WKHLU FRPPXQLFDWLYH YDOXHV DQG WR VXEVHTXHQWO\ ¿QG D correlation between these patterns and other layers of linguistic structure.This paper presents the prosodic pattern analysis of a specific pattern: the excessive prolongation in spontaneous Israeli Hebrew (IH).In order to determine whether it is a random phenomenon or a predictable prosodic pattern, a structural analysis was carried out.First, this phenomenon was taken out of its canonical research framework as a type of speech disfluency (hesitation).Second, its acoustic characteristics were defined, and the phonological environments of these prolongations were accounted for.The inclusion of a prosody-syntax interface in such a framework is also demonstrated.
2. Appended e vowels that are inserted, as enclitics, after a word, but within the same intonation unit, e.g., /az/ 'so' which is pronounced [aze] 'so eh'.Before analyzing the acoustic characteristics of the CE boundary tone, the database on which the research was performed will be introduced.

Spontaneous Hebrew database
The corpus used in this research contains 19 audio segments from 19 recordings that were selected from CoSIH.The recordings, which were made during 2001-2002, are of authentic IH everyday conversations.Each dialogue consists of conversations between one core speaker and various interlocutors with whom the speaker interacted on that day.The research corpus consists of 31,943 words (over 6 hours of speech) of which 4,419 are word-types (unintelligible speech events are not included).All recordings were manually transcribed according to the Speech Assessment Methods Phonetic Alphabet (SAMPA), which is the transcription method used in the present paper too.

The phonetic realization of the CE boundary tone
The phonetic characteristics of the CE boundary tone in IH will be described according to three acoustic parameters (cf.SILBER-VAROD, 2010):  FOX (2000) suggests that the hesitation vowel used by speakers is different even in speakers of different varieties of English.This is also suggested by SHRIBERG (1994), who says that "Filled pauses have variants in many (perhaps all) languages, but their vowel quality varies predictably with the vowel inventory of the language" (ibid, p. 24-25).SHRIBERG (1994) demonstrates how GLVÀXHQF\ LV DOVR DVVRFLDWHG ZLWK DOWHUQDWLRQV LQ YRZHO quality -a prolongated the is pronounced as the variant [Di] (ibid, p. 163).CRUTTENDEN (1997) provides additional examples of the differences in the sound of ¿OOHG SDXVHV EHWZHHQ GLDOHFWV DQG ODQJXDJHV ³,Q 53 DQG LQ PDQ\ RWKHU GLDOHFWV RI (QJOLVK WKH ODWWHU >WKH ¿OOHG pause] typically involves the use of a central vowel [ԥ] ... in Scottish English a sound in the region of the vowel in gate and play is typical while in Russian an alveolar nasal is more common than a bilabial nasal" (CRUTTENDEN, 1997, p. 30).In French, it is the closed-mid front vowel

Phonological environments of CE boundary tones
All possible phonological environments of CE tones are mapped in Table 1, which presents a scheme of the possible syllabic structures that can be found in   The results can be interpreted as the mechanism with which IH speakers manipulate the prolongation.I would argue that this is the same phenomenon, but a different phonological manifestation.Still, it should be mentioned that the two manifestations also occur in other syllable types; i.e. closed syllables are also lengthened, and an appended [e] is also attached to open syllables.This was also found by ROLL, FRID, AND HORNE (2007, p. 229), who measured the durations of the monosyllabic word att µWKDW ¶ LQ 6ZHGLVK ERWK LQ ÀXHQW DQG GLVÀXHQW HQYLURQPHQWV

Syntactic aspects of prolongations
7KH ODVW REVHUYDWLRQ WKDW ZLOO EH PHQWLRQHG EULHÀ\ here is the prosodic-syntactic interface of the excessive prolongation phenomenon.In SILBER-VAROD (2013), it was found that prolongations in IH do not occur randomly, EXW WKDW WKH\ VHUYH D VSHFL¿F FRJQLWLYH IXQFWLRQ 7KLV LV UHÀHFWHG LQ WKH V\QWDFWLF HQYLURQPHQW ZKHUH WKH\ PRVW likely occur in IH as prolongated conjunctions (e.g., [ve] µDQG ¶ SRVVHVVLYH PDUNHUV >6HO@ µRI ¶ GH¿QLWH DUWLFOHV >KD@ 'the', and other function words such as prepositions.This can be explained by mapping the relationship between linguistic and cognitive complexity.It is suggested here, according to GIVÓN (2009), that the ease of production RQ WKH SURVRGLF OHYHO LV UHÀHFWHG E\ ÀXHQW VSHHFK while more coding material is revealed by excessive prolongations.On the syntactic level however, the ease RI SURGXFWLRQ DFWXDOO\ ³UHÀHFWV LWV GHJUHH RI PHQWDO accessibility" (ARIEL, 2001).Function words are known as the most cognitively accessible, according to ARIEL (2001), while content words are the least accessible and consist more of coding material, as GIVÓN ( 2009) says: "...More complex mentally-represented events are coded by more complex linguistic/syntactic structures; … More complex mentally-represented events require more complex mental processing operations.… More complex syntactic structures require more complex mental processing operations."(ibid, 2009, p. 283).This analysis is represented by Fig. 7

Discussion
In the present research several linguistic levels were examined with regard to the single phenomenon of excessive prolongations.First it was analyzed perceptually, after which several acoustic measures were FDUULHG RXW LQ RUGHU WR GH¿QH LWV FKDUDFWHULVWLFV LQ ,+ 2Q WKH SKRQRORJLFDO OHYHO WKH ¿QDOV\OODEOH VWUXFWXUH ZDV studied to determine its correlation to the manifestation of excessive prolongation.Results have shown that there LV D VLJQL¿FDQW FRUUHODWLRQ EHWZHHQ WKH ODVW V\OODEOH structure and the realization of prolongations.Yet, this systematic phonological behavior cannot be the explanation of the linguistic motivation of the prolonged words.For example, the word [rotse] (masculine singular verb form of 'wants') ends with a CV structure as well.Does this mean it will most probably be prolongated?Another word [mayim] 'water' ends with the nasal [m].Does the fact that it is a nasal continuant predict its prolongation over other words?
With the structural multi-level analysis described here, it seems that the phenomenon of excessive prolongations -termed here the CE boundary tone -does not depend on the phonological structure of ZRUGV VLQFH LW KDV EHHQ VKRZQ WKDW WKH ¿QDOV\OODEOH structure does not prevent IH speakers from prolonging syllables, which they also do with an appended [e].Once this regularity was found, the linguistic motivation for excessive prolongations was discussed on the syntagmatic level.It has been shown that, as in other languages, this phenomenon mostly occurs on function words.

Summary and future research
In the present research, three types of prolongations have been assembled into a single prosodic pattern termed the CE boundary tone.Such an approach allows for comparative cross-linguistic studies that examine this perceptual interpretation (e.g. that the speaker wants to continue), universal features (e.g.prolongation), as well as language-dependent features (e.g. the elongated vowel quality).
5HVXOWV DQG DQDO\VLV RI WKH ¿QGLQJV VXJJHVW that excessive prolongations in IH form a pattern which requires parallel segmental and suprasegmental mechanisms.The CE tone, which is mostly perceived as hesitation, is one of the techniques that the speaker uses to signal that he or she has more to say.This prosodic manifestation is executed when phonological structures on the segmental level easily allow it (in open syllables at PRUSKHPH ¿QDO SRVLWLRQ RU ZKHQ SKRQRORJ\ VHHPLQJO\ EORFNV LW DQG WKXV WKH SURVRGLF VWUXFWXUH ³¿QGV LWV ZD\ out" (with an appended [e] vowel).This prosodic pattern is thus a bridge over two intonation units that are syntactically dependent.This mechanism maintains the

ÀRZ RI VSHHFK RU WKH VSHHFK FKDLQLQJ
The structural analysis suggested here can serve as a format for the analysis of other prosodic patterns as well.Like the CE boundary tone, any prosodic pattern can be analyzed for its communicative value, phonetic realization, phonological structure, and syntagmatic environment.
Although a comprehensive analysis was presented above, excessive prolongations in IH still need to be thoroughly investigated on several levels.First, acoustic measurements should be conducted on the entire corpus.Moreover, although results on the syntagmatic level are well studied, the syntactic level analysis should be widened and a segmentation of deep syntactic structures of the corpus should be conducted in order to compare results in other languages and in order to predict excessive prolongations with regard to syntactic complexity.
5HJDUGLQJ WKH VKLIW IURP D GLVÀXHQF\ phenomenon to a prosodic pattern, I would argue KHUH WKDW WKH VDPH SHUFHLYHG GLVÀXHQF\ SKHQRPHQRQ has several phonological manifestations but the same acoustic characteristics.The fact that two manifestations can be considered as a single phenomenon is a linguistic premise, and SHRIBERG (2001) refers to hesitations and ¿OOHG SDXVHV LQ WKLV UHVSHFW WRR "One of the most commonly observed effects RI GLVÀXHQF\ LV D OHQJWKHQLQJ RI UK\PHV RU V\OODEOHV preceding the interruption point.... For example, in the XWWHUDQFH VKRZQ WKHUH LV D UHSHWLWLRQ GLVÀXHQF\ RI µthe the ¶ 7KH ¿UVW LQVWDQFH RI µthe' (which constitutes the reparandum) is much longer than the second instance.... This suggests that in the reparandum, speakers are signaling delay, hesitating PXFK OLNH WKH\ PLJKW GLVSOD\ ZLWK D ¿OOHG pause."(SHRIBERG, 2001, p. 161.My emphasis).

Fig. 1 :
Fig. 1: An example of a syllable under excessive prolongation

Fig. 2 :
Fig. 2: An example of an appended [e] encliticized to a word

Fig. 4 :
Fig. 4: Concatenated mid-front [e] vowels as realized carrying the CE boundary tone in spoken IH [ø], like euh, while in American English, it is the openmid back vowel ‫,]ݞ[‬ like uh or a vowel close to schwa, as mentioned inSHRIBERG (2001, p. 164).
f 0 : A CE boundary is produced on a level-tone (Flat intonation in PORTES AND BERTRAND's (2006) labeling V\VWHP $ OHYHOWRQH LV GH¿QHG DV WKH DYHUDJH NH\ IRU WKH speaker.The measurements of the fundamental frequency of a single female speaker have shown that the tone of the prolongated vowel is 192Hz on average (Fig. 5).

Fig. 5 :
Fig. 5: A level tone of female speaker, as realized at CE boundaries of spoken IH ZRUG¿QDO V\OODEOHV DQG WKH FRUUHVSRQGLQJ H[SHFWHG prolongation type -either syllable prolongation or an DSSHQGHG >H@ 7KH ¿UVW FROXPQ VSHFL¿HV RSHQ YV FORVHG syllable structures.The next column shows the possible ¿QDO VHJPHQWV in the syllable.It is demonstrated in that column that while in closed syllables any consonant can be prolongated in both techniques (prolongation or appended [e]), when it is an open syllable, only four ([i], >D@ >R@ >X@ RXW RI WKH ¿YH YRZHOV LQFOXGLQJ H LQ ,+ FDQ be prolongated with an appended [e].An open syllable ending with [e] cannot be separated, by the listener, for its two /e/ sequences: the word-internal [e] and the appended [e].The prolongated segment column details which segments are expected to undergo prolongation: the vowel in the case of an open syllable, such as in [oÖ] 'or', [keiluÖ] 'as if', [beÖ] 'in'; or an appended e in the case of open syllables that do not end with an e, such as in [ha-e] 'the eh'.In the case of closed syllables, either WKH QXFOHXV YRZHO RU WKH ¿QDO FRQVRQDQW LQ WKH FDVH RI sonorant or continuant codas, may be lengthened.It is also possible for stops to be lengthened, as in the case of WKH 6ZHGLVK GLVÀXHQFLHV att 'that' and och 'and', which were documented and analyzed in ROLL, FRID AND HORNE (2007).Nevertheless, an appended [e] is generally expected following closed syllables, as in [aval-e] 'but eh'.The Measure of duration column indicates which segments should be measured with respect to their duration in each syllabic structure.The Threshold column details the cases where a threshold Fig. 6: 2FFXUUHQFHV RI ZRUG¿QDO V\OODEOH W\SHV SHU SURVRGLF YDULDEOH in the corpus of 230ms is set.It should be noted that in many cases, perceived prolongation were ruled out because of this 230ms threshold.In cases of appended [e]s, no duration measurements should be carried out and appended [e] cases are automatically marked as CE boundaries.
. The two dark gray boxes