FONETIIKAN PÄIVÄT 2006 THE PHONETICS SYMPOSIUM 2006

PUBLICATIONS OF THE DEPARTMENT OF SPEECH SCIENCES UNIVERSITY OF HELSINKI HELSINGIN YLIOPISTON PUHETIETEIDEN LAITOKSEN JULKAISUJA * 53 * FONETIIKAN PÄIVÄT 2006 THE PHONETICS SYMPOSIUM 2006 toim./ed. Reijo Aulanko, Leena Wahlberg & Martti Vainio 2006

Puhetieteiden laitos Department of Speech Sciences Helsingin yliopisto University of Helsinki PL 9 (Siltavuorenpenger 20 A) P.O.Box 9 (Siltavuorenpenger 20 A) 00014 Helsingin yliopisto FI-00014 University of Helsinki ISSN 1795-2425 ISBN 978-952-10-3663-7 (nid./paperback) ISBN 978-952-10-3664-4 (PDF, http://ethesis.helsinki.fi) Hakapaino Oy, Helsinki 2006 Copyright The Authors and the Department of Speech Sciences, University of Helsinki 2006

Fonetiikan päivät 2006 The Phonetics Symposium 2006 iii SISÄLTÖ / CONTENTS Alkusanat... v Foreword... vi Eva Liina Asu: Rising intonation in Estonian: an analysis of map task dialogues and spontaneous conversations... 1 Päivikki Eskelinen-Rönkä: Lentoturvallisuus ja äänitutkimus... 9 Dennis Estill: On Zyryan word stress... 14 Antti Hannukainen, Teemu Lukkari, Jarmo Malinen & Pertti Palo: Formants and vowel sounds by the Finite Element Method... 24 K. Harinen, J. Kirjavainen & A. Iivonen: Segmenttikohtaisen tiedon merkitys puhujien erottelussa... 34 Riina Humalajoki, Maija S. Peltola & Janne Savela: Vieraan kielen vokaalikategorioiden havaitseminen ja oppiminen... 44 Katja Hytönen: Espanjan kielen affrikaatta /ĉ/ suomalaisten ääntämänä: akustis-komparatiivinen tutkimus syntyperäisten mannerespanjan puhujien ja suomalaisten espanjanopiskelijoiden välillä... 48 Annika Hämäläinen, Louis ten Bosch & Lou Boves: Multi-path syllable models based on phonetic knowledge... 57 Antti Iivonen, Jussi Kirjavainen, Beatrice Hauri, Mona Lehtinen & Samuli Saastamoinen: Vieraan kielen sanojen foneettisten muistijälkien vahvistaminen multimediaohjelmalla... 67 Toshiko Isei-Jaakkola: Finnish voice for Japanese speakers: age, gender, and acoustic correlation... 76 Hanna Kärkkäinen, Viola de Silva & Riikka Ullakonoja: Havaintoja perustaajuudesta spontaanin puheen jaksottelussa suomessa ja venäjässä... 84

iv Sisältö / Contents Mietta Lennes, Eija Aho, Minnaleena Toivola & Leena Wahlberg: On the use of the glottal stop in Finnish conversational speech... 93 Einar Meister & Stefan Werner: Intrinsic microprosodic variations in Estonian and Finnish: acoustic analysis...103 Lya Meister: Assessment of the degree of foreign accent: a pilot study...113 Meelis Mihkla: Comparison of statistical methods used to predict segmental durations...120 Marko Moberg & Kimmo Pärssinen: Using text-to-speech in mobile phones...125 Michael O'Dell & Tommi Nieminen: Tahdin ajoitus suomessa oskillaattorimallin näkökulmasta...134 Stina Ojala & Olli Aaltonen: Puheen ja viittomien suhde...144 Tino Ojala: Intelligibility evaluation of present Finnish text-to-speech systems...151 Juhani Toivanen: The fall-rise intonation in the English of highly proficient non-native speakers...161 Juhani Toivanen: Prosodic variation in Finnish-English speech: a corpusbased investigation of two sub-varieties of an L2 speech form...169 Juhani Toivanen, Eero Väyrynen & Tapio Seppänen: Emotion classification from speech in different time windows: human listeners vs. computer...174 Minnaleena Toivola: Äidinkieleltään venäläisten puhujien suomen ääntämisen arviointia...180 Sari Ylinen, Minna Huotilainen & Risto Näätänen: Foneemin kvaliteetin ja kvantiteetin suhde suomenkielisten fonologisessa järjestelmässä...185

Fonetiikan päivät 2006 The Phonetics Symposium 2006 v ALKUSANAT Kahdennetkymmenennetneljännet Fonetiikan päivät pidettiin Helsingissä 30. 31.8. 2006. Tämä suomalaisten puheentutkijoiden kokousperinne sai alkunsa jo 35 vuotta sitten, kun Turussa pidettiin ensimmäiset Fonetiikan päivät toukokuussa 1971. Sittemmin päiviä on järjestetty vuoden tai kahden vuoden välein useissa Suomen kaupungeissa. Suomalaisten puheentutkijoiden ohella virolaiset foneetikot ovat vuosien varrella vakiintuneet osaksi Fonetiikan päivien osanottajia. Vuonna 1998 päivät järjestettiinkin Viron Pärnussa. Nyt pidetyt Fonetiikan päivät olivat esitelmien määrällä mitattuina suurimmat tähänastisista. Päivillä pidettiin yhteensä 39 esitelmää, joista noin puolet järjestettiin nyt ensimmäistä kertaa posterisessioina. Tässä julkaisussa on mukana 24 esitelmää, joiden aihepiirit kattavat laajan alueen ja osoittavat tämänhetkisen puheentutkimuksen monipuolisuuden. Tutkimuskohteina ovat suomen ohella useat muut kielet (englanti, espanja, japani, komisyrjääni, venäjä, viro ja viittomakieli) ja hyvin monenlaiset puheenkäyttötilanteet. Fonetiikan sovelluskohteina käsitellään mm. kielenopetusta, puheteknologiaa ja forensista tutkimusta. Useimpien muiden pienten yliopistollisten aineiden tavoin fonetiikka on viime vuosina saanut tuntea olevansa ahtaalla erilaisten taloudellisten ja tiedepoliittisten linjausten takia. Valtionhallinnon ns. säästötoimet kohdistetaan usein sokeasti pieniin oppiaineisiin ja tieteenaloihin pohtimatta tarkemmin näiden todellista yhteiskunnallista merkitystä. Nyt järjestettyjen Fonetiikan päivien aihealueiden monimuotoisuus osoittaa, että fonetiikka on Suomessa elinvoimainen ja kehittyvä ala. Haluamme kiittää Helsingin yliopistoa ja Suomen Akatemiaa taloudellisesta tuesta. Kiitos päivien käytännön järjestelyjen onnistumisesta kuuluu ennen kaikkea uhrautuvalle talkooväelle (Jussi Kirjavainen, Mietta Lennes, Minnaleena Toivola) ja opiskelija-avustajille (Mona Lehtinen, Rina Pitkänen). Lopuksi kiitämme kaikkia päiville osallistuneita mielenkiintoisen ja antoisan tieteellisen kokouksen aikaansaamisesta. Omistamme tämän julkaisun dos. Antti Iivoselle, pitkäaikaiselle Helsingin yliopiston fonetiikan professorille, joka käynnisti näiden Fonetiikan päivien järjestelyt ja vaikutti monin tavoin päivien onnistumiseen. Helsingissä 22.12.2006 toimittajat

vi Fonetiikan päivät 2006 The Phonetics Symposium 2006 FOREWORD The 24 th Finnic Phonetics Symposium was organised in Helsinki at the end of August 2006. The tradition of the Meetings for Finnish phoneticians began 35 years ago, as the first meeting was held in Turku in May 1971. Since then the meetings have been organised annually or biannually in different towns in Finland. In addition to Finnish speech scientists, Estonian phoneticians have been an integral part of the meetings for a long time. The English term "Finnic Phonetics" was adopted in 1998 when the meeting was organised in Pärnu, Estonia. With regard to the number of papers, the current meeting in Helsinki was the biggest so far. Altogether 39 presentations were given, about half of them as posters. This publication contains 24 papers the topics of which cover a wide area and show the multifaceted nature of speech research. The topics include a wide variety of languages (Finnish, English, Estonian, Japanese, Russian, Spanish, Zyryan and the Finnish sign language) and many different speaking conditions and styles. The authors apply phonetic research to, e.g., language teaching, speech technology and forensic purposes. Like most other small academic disciplines, phonetics has felt the pressure from economic and science policies, which do not take into account the real impact of these disciplines to the society. The current Symposium with its wide variety of topics proves, however, that phonetics is a viable and advancing discipline in Finland. We would like to thank the University of Helsinki and the Academy of Finland for financial support. The smooth organisation of the actual meeting was made possible by our dedicated voluntary personnel (Jussi Kirjavainen, Mietta Lennes, Minnaleena Toivola) and our student helpers (Mona Lehtinen, Rina Pitkänen). Finally, we extend our thanks to all the participants for making the Symposium an interesting and fruitful scientific meeting. We dedicate this publication to docent Antti Iivonen, a long-standing professor of phonetics at the University of Helsinki, who set in motion the organising of this Symposium and contributed in many ways to its success. Helsinki, December 22, 2006 the editors

Fonetiikan päivät 2006 The Phonetics Symposium 2006 1 RISING INTONATION IN ESTONIAN: AN ANALYSIS OF MAP TASK DIALOGUES AND SPONTANEOUS CONVERSATIONS Eva Liina Asu Institute of the Estonian Language, Tallinn, Estonia, ela@cantab.net Abstract This paper examines the distribution, phonetic realization, and discourse function of rising tunes in Estonian map task dialogues and spontaneous conversations. The analysis shows that, contrary to traditional accounts, rises in Estonian conversational data are very common. Rising intonation seems to be most commonly associated with short discourse markers such as jaa, mhmh, noo, the main function of which is to mark recipiency and signal the interlocutor s orientation to communication. Rising intonation also occurs elsewhere both in turn-final and turn-medial positions. It is possible to distinguish two different phonetic shapes, concave and convex, which fall broadly within two phonological categories. A closer examination of the discourse functions reveals that the different F0 contours imply a difference in communicative meaning: a rise with a high boundary carries a turn yielding meaning whereas the rise ending in a plateau is more likely to act as a turn holder. Keywords: rising tunes, intonational phonology, communicative functions, phonetic realization, discourse markers, boundary tones 1 Introduction The existence of rises in Estonian has been commented on since the earliest experimental studies on Estonian intonation (e.g. Peters 1927) but there seems to be confusion as to what the attitude towards rising tunes should be: are they part and parcel of the Estonian intonational inventory, or are they something strange and rare? In traditional accounts of Estonian phonetics (e.g. Ariste 1977) and in prescriptive textbooks (e.g. Kraut et al. 1998), all rising intonation is labeled as uncommon and foreign sounding. In a commentary for Estonian language teachers accompanying a textbook, Ehala (1999: 16) describes conservative Estonian intonation as traditionally monotonous with a fall at the end of a sentence. Such accounts have relied almost exclusively on the study of read speech, which in itself is not surprising as until recently very little research had been carried out on the intonation of spontaneous Estonian. Lately, there has been a noticeable shift in focus of study from read to spontaneous speech although little research still exists on the intonation of spontaneous Estonian. Interestingly, the few small-scale studies available suggest that rising intonation might be much more common than previously thought.

2 Eva Liina Asu: Rising Intonation in Estonian A study by Asu and Nolan (2003), which tests the preliminary phonological inventory of Estonian intonation based on read speech against the phonology of spontaneous speech, finds a high degree of correspondence between the two types of data in terms of pitch accents used, but shows that the spontaneous data contains considerably more rising intonation patterns than the read one. A study by Keevallik (2003) also reports a large number of terminal rises in Estonian conversational data which consists of recorded phone conversations. The study focuses on rising response tokens by the interlocutor which are shown to signal continuation on the part of the speaker and function as a kind of go-ahead. Keevallik (2003: 62) concludes that in order to account for all the prosodic phenomena in a language it is necessary to look at interactional sequences, and therefore rising contours in Estonian can only be studied in the context of interaction. Likewise, research by Ogden and Routarinne (2005) on rising intonation in Finnish implies that it is fruitful to combine the study of intonation with the study of interaction. Their work investigates the occurrence and communicative functions of final rises in conversations between teenage girls from Helsinki and shows that rising intonation in their data signals primarily a place where the coparticipant in the conversation can mark recipiency, or else it indicates continuation by showing that more talk is to follow by the same speaker. An autosegmental-metrical (AM) account of Estonian intonation (Asu 2004) distinguishes between two different types of rises according to the point at which the pitch rises and the nature of the intonational phrase (IP) boundary. In the first type, a rise starts on or immediately after the accented syllable and is followed by a high plateau. In the AM notation adopted for the Estonian language, such rises are labeled with an L*+H pitch accent which is followed by an unmarked IP boundary (0%). In the second type, a rise from a low accented syllable L* is followed by a high IP boundary (H%). Variation in the shape of rising contours has also been shown for other languages such as for instance German. Dombrowski and Niebuhr (2005) describe phrase-final rises on a convex-concave continuum. The phonetic variation of these patterns triggers different conversational conditions: a contour curved outwards results in the turnholding condition; a contour curved inwards results in the turn-yielding condition. Other than Keevallik s (2003) study on response tokens there is virtually no research carried out on the interactive functions of intonation in spontaneous Estonian. Therefore, the goal of the present paper is to look at Estonian intonation in interaction. More specifically, the aims are to examine rising tunes in spontaneous Estonian by analyzing their distribution, phonetic characteristics and discourse function. The research presented will seek answers to the following questions: How common is rising intonation in Estonian conversational data? Is there confirmation of the previously postulated two phonological categories of rises? Is there a correlation between the type of rise and the response type from the interlocutor? 2 Materials and method The materials used for the current study consist of two sets of data: map task dialogues and spontaneous conversations. The map task method (Brown et al. 1984) was originally developed for studying communication and collaboration strategies and language use of different speakers. It is a cooperative task involving two participants, one of whom is designated as Instruction Giver and the other as Instruction Follower. Each of the two speakers has a map which the other one cannot see: the Instruction

Eva Liina Asu: Rising Intonation in Estonian 3 Giver s map has a route marked on it whereas the Instruction Follower s has not. The task of the speakers is to reproduce the route of the Instruction Giver s map on the Instruction Follower s map. The map task data used here comprises six map task dialogues recorded by six young female speakers of Standard Estonian. The participants who were familiar with each other beforehand worked in pairs. Each pair was recorded while performing two map tasks so that both members of the pair could take the role of Instruction Giver and Instruction Follower. After finishing the first task the roles were swapped while each speaker was given a different version of the map she had used in the first task. The material from both roles was included in the analysis. The total duration of the map task data is 27 minutes. Spontaneous conversations were recorded between the same pairs of speakers during the same recording session as the map tasks. The speakers were asked to interview each other on such general topics as summer holidays and food. The total duration of the conversational data is 56 minutes. All the recordings were conducted in a quiet environment in Tartu, Estonia in the summer of 2002, using a Sony TCD D8 portable DAT recorder and two AKG D190E dynamic microphones with a cardioid response. The microphones were attached to the recorder via an amplifier and placed on stands at about 30 cm from the speakers mouths. The recorded speech data was redigitized at 16 khz using Xwaves+ running on a Silicon Graphics Unix workstation. The analysis was carried out using the speech analysis software Praat (Boersma & Weenink 2004). The data was transcribed orthographically and labeled prosodically. All instances of high terminals were identified and classified according to the realization of the F0 contour, following the classification of Asu (2004) which distinguishes between rises ending in a plateau and rises with a high IP boundary. In order to quantify the phonetic shapes of the rises a small set of comparable disyllabic words was chosen and measurements were taken at five measurement points of the contours: (1) the beginning of the first vowel begv1, (2) the midpoint of the first vowel midv1, (3) the end of the first vowel endv1, (4) the beginning of the second vowel begv2, (5) the end of the second vowel endv2. Finally, to study the communicational functions of rises response or lack of response to the rising contour by the interlocutor was observed and noted down and correlated with the two different contour shapes. 3 Results and discussion The following subsections will present the findings in the order of the questions posed in the introductory section and discuss the distribution, phonetic realization and discourse function of rising intonation in Estonian. 3.1 General distribution of rising intonation Both the map task data and the conversation data contained an abundance of rising tunes. Most commonly, rising intonation was observed to occur on discourse markers (or response tokens) which in Estonian are normally monosyllabic words such as jaa, jah, ah, mhmh, mm ( yes/yeah ), noo ( okay/yeah ), nii ( so/okay ), etc. In spontaneous dialogues, the rate of occurrence of such rising response tokens was 1.8 per minute (i.e. approximately one in every 34 seconds). In the map task data, rises on response tokens were even more frequent occurring at a rate of 3.5 per minute (i.e. one in every 17 seconds).

4 Eva Liina Asu: Rising Intonation in Estonian Figure 1 presents the F0 traces and waveforms of the two most commonly appearing response tokens from the map task data. The rise is realized towards the end of the syllable ending in a high boundary (L* H%). In the data, rises on such short response tokens were often additionally preceded by a downward onglide to the low accent as can be seen (albeit in a narrow pitch span) in the first example (jaa). jaa yeah noo okay Figure 1. F0 traces and waveforms of the response tokens jaa yeah and noo okay. The following dialogue extract illustrates a typical conversational sequence from the map task data. Speaker A acts as the Instruction Giver explaining the route on her map to Speaker B, the Instruction Follower. All turns by Speaker B in this extract consist only of the response token noo/njoo (= okay ) uttered with rising intonation (L* H%). The square brackets in turns (4) and (5) indicate overlapping speech. (1) A: ja siis tuleb keerata järsult tagasi millegi pärast and then you have to turn abruptly back for some reason (2) B: noo (L* H%) yes/okay (3) A: selle pärast, et seal on mingisugused kalurid because there are some kind of fishermen there (4) B: [njoo] (L* H%) okay/so (5) A: [mingi]sugused mehed some kind of men (6) B: noo (L* H%) okay The main discourse function of such response tokens seems to be the marking of recipiency and giving feedback. By providing a response the interlocutor is signaling

Eva Liina Asu: Rising Intonation in Estonian 5 her orientation to the utterance, or giving the so called go-ahead (Keevallik 2003). In addition to discourse markers terminal rises also occurred elsewhere in the data although very rarely on utterances functioning as questions. On the whole, rising intonation was five times more frequent in the map task data than in the spontaneous dialogues. This can be explained by the nature of the map task data as such, as the task is inherently communicational and requires collaboration on the part of the participants. The communicational nature of the map task is also reflected in the fact noted above that discourse markers were twice as common in the map task data than in the conversational data (3.5 vs 1.8 per minute). 3.2 Phonetic realization of rises When looking at the phonetic realization of rises two distinct F0 shapes could be distinguished. These fall broadly within the two phonological categories which were previously postulated by Asu (2004). A rise can end with a high IP boundary (H%) as was the case with most short response tokens in the map task data, or it can plateau out ending with an unmarked boundary (0%). Figure 2 exemplifies these two different phonetic shapes of final rises on the same segmental material the postposition alla ( down ). The panel on the left shows the concave contour shape, and the panel on the right the convex shape. Concave shape (alla down ) Convex shape (alla down ) Figure 2. F0 traces and waveforms exemplifying the two different phonetic realizations of rising phrase final contours To quantify the different F0 shapes the measurements described above were carried out on a set of words which occurred in the data in both phonetic realizations a comparable number of times. Obviously, the number of tokens was very limited (due to it being uncontrolled spontaneous data) but at least enables us to compare the different realizations of the F0 in a more controlled way. The quantification of the contour shapes is presented in Figure 3 using the word mööda ( along/by ) which acts as a postposition and occurred frequently in the data in phrase-final position. As can be seen, the rise in the convex shape starts immediately after measurement point 1 at the beginning of the contour (begv1) and continues until point 4 (begv2) where it reaches a plateau which stays level until point 5 (endv2). In the concave shape, the initial contour is more or less level until measurement point 3 (end V1) where the rise starts. The rise continues through point 4 (begv2) until the very end of the contour (endv2).

6 Eva Liina Asu: Rising Intonation in Estonian pitch (st) 25 20 15 10 5 0 1 2 3 4 5 6 measurement points Figure 3. Mean contours of concave and convex rises on the word mööda ( along/by ) 3.3 Correlation of contour shape and discourse function As demonstrated by Dombrowski and Niebuhr (2005) for German, phonetic variation in contour shape can trigger different conversational conditions. The analysis of responses to the different rising patterns in the present data also seems to reveal a correlation between the contour shape and discourse function. It appears that the concave shape is followed by a turn change in 55% of the cases, which suggests that this contour shape is more likely to be directed towards the interlocutor and carry a turn-yielding function. The convex shape, on the other hand, is followed by a turn change in only 17% of the cases, which strongly implies an orientation toward the speaker and suggests that the contour carries a turn-holding function. The following extract from the map task data briefly illustrates the different communicative functions of the two contour shapes. In turn (1) Speaker A, the Instruction Giver, is explaining a route which involves walking along a grove of trees and down by the side of a pine tree. While giving her instructions she twice uses the convex contour (on the words alla ( down ) and mänd ( pine )) signaling that there is more talk to follow. She ends this turn with a falling intonation. Turn (3), however, ends with a concave shaped rise, which could indicate (similarly to falling intonation) that she has finished talking and is inviting feedback from the interlocutor. Speaker B responds both in turns (2) and (4) with the response token mhmh (= yeah ) realized as in the example above with the rise ending with a high IP boundary. (1) A: puude juurest ida küljest alla (L*+H 0%) siis tuleb mänd (L*+H 0%) ja männist läänepoolsest küljest alla near the trees along the eastern side down then comes the pine and along the western side of the pine down (2) B: mhmh (L* H%) yeah (3) A: teeb s-tähe ja talu eest läbi (L* H%) (it) makes the letter s and passes in front of the farm house (4) B: mhmh (L* H%) yeah

Eva Liina Asu: Rising Intonation in Estonian 7 The present results are in line with those of Dombrowski and Niebuhr (2005) for German showing evidence for there being a correlation between the phonetic form and discourse function. Similarly to, for instance, English discourse where the function of rises is first and foremost thought to be cooperative (e.g. Ladd 1996), rising intonation in Estonian (even if rare in read speech) seems to play an important interactional role in a conversational setting (e.g. Asu & Nolan 2003). 4 Conclusions This paper studied the distribution and phonetic characteristics as well as discourse functions of rises in a database of Estonian spontaneous speech elicited with the help of the map task and spontaneous conversations between six young female speakers of Standard Estonian. The analysis showed that contrary to the traditional accounts of Estonian intonation but consistent with some earlier small-scale studies on conversational speech, rising intonation in spontaneous Estonian is very common. In map task dialogues, for instance, rises on monosyllabic discourse markers occurred at a rate of 3.5 per minute. There were also plenty of phrase-final rises elsewhere in turn final and turn-medial positions. Two different phonetic shapes (convex and concave) were distinguished, which correspond to two phonological categories proposed for Estonian in Asu (2004). An analysis of the correlation between the contour shape and discourse function shows that different rising contours seem to imply a difference in communicative intent. This is consistent with earlier findings for other languages such as for instance German, and Finnish. The inwards-curving concave shape (a rise with a high boundary) tends to carry an interactional meaning. It is either turn yielding, signaling a place for an interlocutor to mark recipiency, or else it can be used for marking feedback as in the response tokens. At the same time, a rise ending in a plateau acts more as a continuation marker. Its function tends to be turn holding, projecting more talk by the current speaker. 5 Acknowledgement While carrying out the work presented in this paper the author was funded by the Estonian Science Foundation post-doctoral grant (project no. 0052488s03). 6 References Ariste, Paul (1977). Eesti Keele Foneetika. Tartu: Tartu Riiklik Ülikool. Asu, Eva Liina (2004). The Phonetics and Phonology of Estonian Intonation. University of Cambridge: doctoral dissertation. Asu, Eva Liina & Nolan, Francis (2003). Testing the Model of Estonian Intonation. In M. J. Sole, D. Recasens & J. Romero (Eds.), Proceedings of the 15th International Congress of Phonetic Sciences (pp. 1249 1252). Barcelona. Boersma, Paul & Weenink, David (2004). Praat: doing phonetics by computer (Version 4.1.13) [Computer program]. Retrieved August, 2004, from http://www.praat.org/ Brown, Gillian, Anderson, Anne, Shillcock, Richard & Yule, George (1984). Teaching Talk: Strategies for Production and Assessment. Cambridge: Cambridge University Press. Dombrowski, Ernst & Niebuhr, Oliver (2005). Acoustic Patterns and Communicative Functions of Phrase-Final F0 Rises in German: Activating and Restricting Contours. Phonetica 62, 176 195. Ehala, Martin (1999). Eesti Kirjakeel. Gümnaasiumi õigekeelsusõpik. Tallinn: Künnimees.

8 Eva Liina Asu: Rising Intonation in Estonian Keevallik, Leelo (2003). Terminally Rising Pitch Contours of Response Tokens in Estonian. Crossroads of Language, Interaction, and Culture, Vol. 5, 49 65. Kraut, Einar, Liivaste, Ene & Tarvo, Aili 1998. Eesti Õigekeel. Tallinn: Koolibri. Ladd, D. Robert (1996). Intonational Phonology. Cambridge: Cambridge University Press. Ogden, Richard & Routarinne, Sara (2005). The Communicative Functions of Final Rises in Finnish Intonation. Phonetica 62, 160 175. Peters, William (1927). Esimene katse eesti keele kõnemeloodia võrdlevas uurimises [part 2]. Eesti keel 6(1-3), 1 48.

Fonetiikan päivät 2006 The Phonetics Symposium 2006 9 LENTOTURVALLISUUS JA ÄÄNITUTKIMUS Päivikki Eskelinen-Rönkä Helsingin yliopisto, puhetieteiden laitos paivikki.eskelinen-ronka@pp.inet.fi Tiivistelmä Suomen Kulttuurirahaston yhteydessä toimivan Etelä-Savon rahaston tuella on käynnistetty tutkimushanke, jonka tavoitteena on kehittää uusia tutkimusmenetelmiä ilmailuonnettomuuksien tutkintaan. Uudet menetelmät perustuvat kone- ja aerodynaamisten äänien tutkimukseen. Ohjaamoäänitallenteet sisältävät puheen lisäksi runsaasti erilaisille lentotiloille tyypillisiä koneääniä ja aerodynaamista kohinaa. Sisällyttämällä myös tämä ääniaineisto tutkimuksiin voidaan äänitallenteiden tutkinnallista hyödynnettävyyttä tehostaa merkittävästi onnettomuuden tai vaaratilanteen syyn selvityksessä. Tutkimusmenetelmien monipuolistuminen tuo lisää tietoa lentoturvallisuuteen vaikuttavista tekijöistä. Tutkimuksella on siis oma vaikutuksensa myös lentoturvallisuuden parantamiseen. Kehitettävillä menetelmillä on sovellusalue myös rikostutkinnassa silloin, kun tutkittavana on ilmailun radiopuhelinliikenteen häirintä tai laiton puuttuminen ilma-aluksen kulkuun. Keywords: Onnettomuustutkinta, ohjaamoäänitallenteet, äänitutkimus, lentoturvallisuus 1 Johdanto Suomessa onnettomuuksien tutkinnasta vastaa oikeusministeriön alainen Onnettomuustutkintakeskus. Onnettomuustutkinnan tehtävänä on selvittää onnettomuuteen tai vaaratilanteeseen johtaneet ja/tai myötävaikuttaneet tekijät siten kuin laissa ja säädöksissä on määrätty (Laki onnettomuuksien tutkinnasta, Asetus onnettomuuksien tutkinnasta). Tutkinta ei ota kantaa rikosoikeudellisiin syyllisyys- tai vastuukysymyksiin. Valmistuneet tutkintaselostukset ovat julkisia ja ne ovat luettavissa verkkosivuilla osoitteessa http://www.onnettomuustutkinta.fi/. 2 Lentotietojen tallennus Liikenneilmailua harjoittavat ilma-alukset on kansainvälisten määräysten mukaisesti varustettu tallenninjärjestelmillä, jotka rekisteröivät useita lennon aikana tapahtuvia toimintoja. Taustalla tapahtuva tallentaminen on luonteeltaan automaattista ja järjestelmien käyttäjille huomaamatonta. Kansainvälisten sopimusten mukaan tallennuslaitteet ovat ulkoiselta väritykseltään kirkkaan oransseja. Kuorirakenteeltaan ne täyttävät iskun-, paineen- ja palonkestävyyttä koskevat standardit (EUROCAE 2003). Arkipuheessa näitä laitteita kutsutaan mustiksi laatikoiksi.

10 Päivikki Eskelinen-Rönkä: Lentoturvallisuus ja äänitutkimus Tallentamisen merkitys nousee korostetusti esille silloin, kun lentoliikenteessä on syntynyt vakava vaaratilanne tai tapahtunut onnettomuus (Kuva 1). Näin siksi, että tallennuslaite sisältää ilma-aluksen fyysisen muistin onnettomuuteen johtaneesta tapahtumaketjusta. Kuva 1. Nokkatelineen toimintahäiriö Helsinki-Vantaan lentoasemalla 8.4.2001 (Lähde: http://www.onnettomuustutkinta.fi/uploads/2iloh28vgi.pdf) Onnettomuustutkinnassa mustan laatikon sisältämää muistitietoa käytetään yhdessä muista lähteistä ja erillistutkimuksista saatujen onnettomuutta koskevien tietojen kanssa (Kuva 2). Tietojen yhdistämisellä muodostettu kokonaiskuva tapahtuneesta paljastaa usein, mitkä tekniset ja/tai inhimilliset tekijät joko aiheuttivat onnettomuuden tai myötävaikuttivat sen syntyyn. Huolellisesti analysoitua tietoa voidaan jatkossa käyttää turvallisempien lentomenetelmien ja teknisten ratkaisujen kehittämiseen. Kuva 2. Tallentimen muistin purkaminen 3 Musta laatikko mustat laatikot Nimitystä musta laatikko voidaan käyttää kahdesta tehtävältään erilaisesta tallentimesta. Lentoarvotallennin (FDR, Flight Data Recorder) kerää lentoon liittyvää teknistä tietoa, kuten moottoriarvoja, lentokorkeuksia, ohjaussuuntia, nopeuksia ja ohjainten liikkeitä. Ohjaamoäänitallennin (CVR, Cockpit Voice Recorder) puolestaan tallentaa ohjaajien keskinäisen sekä ohjaajien ja lennonjohdon välisen kommunikoinnin lisäksi kaikki ohjaamotilassa ja pienemmissä ilma-aluksissa osin myös matkustamosta kuuluvat äänet. Mustien laatikkojen tallennusmenetelmä perustui aiemmin analogitekniikkaan. Tekniikan kehittymisen myötä tallennusjärjestelmissä ollaan kuitenkin vähitellen siirtymässä digitaalitekniikkaan. Uusi tekniikka mahdollistaa 30 minuutin jatkuvapyöritteistä nauhaa huomattavasti pitemmät tallennusajat (30 min 2 h). Digitaalisen tallennusjärjestelmän toiminta ei myöskään häiriinny lentotilojen voimakkaiden muutosten vuoksi yhtä helposti kuin liikkuvia osia sisältävän analogisen

Päivikki Eskelinen-Rönkä: Lentoturvallisuus ja äänitutkimus 11 tallentimen toiminta. Näitä digitaalisia tallentimia kutsutaankin nimellä solid-statetallentimet. Ohjaamoääni- ja lentoarvotallentimet voivat olla yhdistettynä samaan kuoreen (SSCVFDR, Solid State Combined Voice and Flight Data Recorder) tai ne voivat olla kokonaan erilliset tallentimet. Edellä mainittujen tallentimien lisäksi monissa liikennekoneissa on niin kutsuttu QAR-tallennin (Quick Access Recorder). Parametritietoja huollon ja lentoyhtiöiden tarpeisiin keräävä QAR-tallennin ei kuitenkaan ole suunniteltu kestämään onnettomuustilanteen ääriolosuhteita, joten sen sisältämät tiedot eivät välttämättä ole käytössä tutkinnassa. 3.1 Lentotietojen tutkinta käytännössä Kuinka lentoarvo- ja ohjaamoäänitallenteiden tutkimus sitten tapahtuu? Puhetta sisältävien ohjaamoäänitallennusten perusteella pyritään saamaan selville vaaratilanteen tai onnettomuuden välitön tai myötävaikuttava syy sellaisena kuin ohjaaja/ohjaajat sen havaitsevat (Nevile 2005). Kuvassa 3 on esimerkki ohjaamoäänitallentimen sisältämästä ääni-informaatiosta. Tallennus sisältää puhetta eri kanavilta sekä erityisen synkronointisignaaliin. Tarkasti määritetyn synkronointisignaalin avulla voidaan yhdistää sekä CVR:n eri kanavien äänitietoa toisiinsa että CVR:n äänitietoa edelleen FDR:n lentoparametreihin. Kuva 3. Esimerkki CVR:n sisältämästä informaatiosta Lentoarvotallentimeen kirjautuneista toiminnallista parametreista tutkitaan lentotilassa tapahtuneet muutokset. Kuvassa 4 on esimerkki FDR-tallenteesta valittujen parametrien visualisoinnista. Yhdistämällä puheesta ja teknisistä parametriarvoista saatu tieto voidaan ihannetapauksessa esittää selkeä kuvaus tapahtumaketjun etenemisestä sekä siihen vaikuttavista eri tekijöistä. Edellä kuvattu tutkintaideaali toteutuu kuitenkin käytännössä melko harvoin. Ohjaamoäänitallenteen sisältämä puhe voi olla ymmärrettävyydeltään heikko lentotilan vuoksi (esim. g-voimat) tai ohjaamotilan osittain jatkuvan ja tasaisen sekä epänormaalissa tilanteessa mahdollisesti nousevan melutason vuoksi (Keränen et al. 2004, Wagstaff 2001, Hansen et al. 2000, Begault & Wenzel 1998, Tobias 1968a & b). Hätätilanteessa puhekommunikaatio voi jäädä määrällisesti vähäiseksi myös puutteellisen ohjaamoyhteistyön vuoksi (Crew Resource Management, CRM). Kun puhe puuttuu tai on vähäistä, saattaa se vaikeuttaa myös FDR-parametritietojen perusteella tehtävää analyysia tilanteen alkuun panneesta syystä. Onnettomuuteen tai vaaratilanteeseen johtaneen tapahtumaketjun alkusyy voi kuitenkin olla kuultavissa tallennusten muussa kuin puhutussa äänitiedossa (vrt. Forrester 1996). Tallenteen sisältämästä ääni-informaation kokonaismäärästä tällaisen aineiston osuus on huomattavasti suurempi kuin puheaineiston.

12 Päivikki Eskelinen-Rönkä: Lentoturvallisuus ja äänitutkimus Kuva 4. Esimerkki FDR:n lentoparametrien visualisoinnista (Lähde: http://www.aaib.dft.gov.uk/sites/aaib/publications/formal_reports/1_2005_g_bjv x.cfm) 4 Tutkimuskohteena ohjaamon tekniset äänet Autenttiset CVR-ohjaamoäänitallenteet eivät ole julkista tutkimus- tai tutkinta-aineistoa. Sisältöä koskevien suojaustoimenpiteiden vuoksi myös puhe- ja ääniaineiston tutkimusmenetelmiä käsittelevä kirjallisuus on ollut vähäistä (vrt. Eskelinen-Rönkä 2005). Suomessa tallenteiden tutkinnassa on hyödynnetty kokeneiden ja mahdollisuuksien mukaan saman tyyppikoulutuksen saaneiden lentäjien tietotaitoa. Kuunteluun ja ammattikokemukseen perustuva tutkimustapa ei kuitenkaan ole aina ongelmaton (Beringer & Harris Jr. 2005, Kuronen 2004). Siksi tämän kvalitatiivisen tutkimuksen rinnalle tarvitaan kvantitatiivisia eli mittaustietoon perustuvia menetelmiä. Puheen ohella ohjaamoäänitallenteet sisältävät runsaasti eri lentotiloille tyypillisiä teknisiä ääniä ja aerodynaamista kohinaa. Tämän materiaalin systemaattinen hyödyntäminen onnettomuustutkinnassa on ollut vähäistä (Eskelinen-Rönkä & Niemi- Laitinen 2001). Ohjaamoäänitallennusten sisällöllinen erityislaatu tekee kvantitatiivisten tutkimusmenetelmien kehitystyön monella tapaa haastavaksi. Luotettavaa tietoa ei ole saatavilla esimerkiksi siitä, missä määrin yleisesti käytettävissä olevat mittausmenetelmät soveltuvat taustahälyihin sekoittuneiden teknisten äänien tutkimukseen. Tutkimusmenetelmien kehittäminen edellyttää myös hyvin dokumentoitua ja monipuolista mallinäytekokoelmaa eri ilma-alusten ohjaamoäänistä normaalitilanteessa. Tällaista äänitietokantaa ei kuitenkaan ole saatavilla. Ellei koneyksilön sisäistä ja yksilöiden välistä vaihtelua koskevaa tietoa ole käytettävissä, voi yksittäisen mittaustulosten tutkinta johtaa esimerkiksi koneen vikaantumista koskevaan virhepäätelmään. Etelä-Savon rahaston tuella käynnistetyssä hankkeessa aloitetaan eri ilma-alustyyppien ohjaamoäänien tietokantatallennukset sekä koneäänien ja aerodynaamisten äänien tutkimukseen soveltuvien menetelmien kehittely. Muodostettavaan äänitietokantaan kerätään erilaisissa konetyypeissä esiintyvät normaalit ohjaamoäänet eri lennonvaiheessa. Tietokanta ja ohjaamoäänten tutkimukseen kehitettävät menetelmät tulevat olemaan Onnettomuustutkintakeskuksen käytettävissä.

Päivikki Eskelinen-Rönkä: Lentoturvallisuus ja äänitutkimus 13 5 Lähteet Begault, D. R. & Wenzel, E. M. (1998). Assessment of Noise Exposure in Commercial Aircraft Cockpits. Ames Research Center, Moffett Field, CA: Human Information Processing Research Branch. Beringer, D.B. & Harris, H.C. Jr. (2005). A Comparison of Baseline Hearing Thresholds Between Pilots and Non-Pilots and the Effects of Engine Noise. Civil Aerospace Medical Institute Federal Aviation Administration. DOT/FAA/AM-05/12, Office of Aerospace Medicine Washington, DC 20591 Eskelinen-Rönkä, P. (2005). Intelligibility of Radio Communication in Civil Aviation. Academic Dissertation. Publications of the Department of Speech Sciences, University of Helsinki, 51. Eskelinen-Rönkä, P. ja Niemi-Laitinen, T. (2001). The Effect of Noise in Forensic Speech and Audio Analysis. Teoksessa T. Salmi-Tolonen & R. Foley & I. Tukiainen (toim.) Proceedings of The Law and Language - Prospect and Retrospect Congress, Levi 12.- 15.12.2001. CD-Rom. EUROCAE (2003). Minimum Operational Performance Specification for Crash Protected Airborne Recorder Systems, ED-112. France: The European Organization for Civil Aviation Equipment. Forrester, B. D. (1996). Advanced Vibration Analysis Techniques for Fault Detection and Diagnosis in Geared Transmission Systems. Academic Dissertation. Swinburn University of Technology. Australia. Hansen, J.H.L., Swail, C., South, A.J., Moore, R.K., Steeneken, H., Cupples, E.J., Anderson, T., Vloeberghs, C.R.A., Trancoso, I. & Verlinde, P. (2000). The Impact of Speech Under `Stress' on Military Speech Technology. Nato Project 4 Report. Nato Ist/Tg-01. Keränen, H., Väyrynen, E., Pääkkönen, R., Leino, T.,Kuronen, P., Toivanen, J. & Seppänen, T. (2004). Prosodic Features of Speech Produced by Military Pilots during Demanding Tasks. Teoksessa T. Seppänen, K. Suomi ja J. Toivanen (toim.) Fonetiikan päivät 2004 The Phonetics Symposium 2004 (s. 88 91). Oulu: Oulun yliopistopaino Kuronen, P. (2004). Military Aviation Noise. Noise-induced hearing impairment and noise protection. Academic Dissertation. Department of Otorhinolaryngology, University of Oulu. Nevile, M. (2005). Communication in Context: A Conversation Analysis Tool for Examining Recorded Voice Data in Investigations of Aviation Occurrences. ATSB research and analysis report B2005/0118, Final. Tobias, J. V. (1968a). Cockpit Noise Intensity: Fifteen Single-Engine Light Aircraft. Civil Aerospace Medical Institute Publications. CAMI Publications Technical Reports. Report No. FAA-AM-68-21. Tobias, J. V. (1968b). Cockpit Noise Intensity: Eleven Twin-Engine Light Aircraft. Civil Aerospace Medical Institute Publications. CAMI Publications Technical Reports. Report No. FAA-AM-68-25. Wagstaff, A. S. (2001). Speech Intelligibility and Noise in Aircraft. Academic Dissertation. Royal Norwegian Air Force Institute of Aviation Medicine. Departmentgroup for Clinical Medicine, Faculty of Medicine, University of Oslo. Norway, GCS Multicommunication AS, Oslo: Unipub forlag, Akademika AS.

14 Fonetiikan päivät 2006 The Phonetics Symposium 2006 ON ZYRYAN WORD STRESS Dennis Estill University of Helsinki, dennis.estill@helsinki.fi Abstract There are two Komi literary languages and a third Komi dialect that could be considered a separate language. The language under discussion in this paper is Komi-Zyryan (henceforth Zyryan). Since little research has so far been carried out this is a first attempt to embark on this subject by first of all considering Zyryan word stress. The research is in two parts: an auditory experiment followed by an acoustic analysis of the speech of four individuals. In the acoustic analysis, I have considered three parameters; viz. intensity at three levels A 0, A 1, A 3, fundamental frequency (F 0 ) and duration. Voice quality was not analysed. The results, which are still in the process of analysis and so are by no means final, appear to show that in Zyryan word stress is achieved by an increase in intensity, a rise in fundamental frequency, an extension of duration, or any combination of all three. Intensity is a somewhat more frequent cause of word stress. Nevertheless, there do seem to be a proportionally large number of instances in which no relevant stress-causing parameter can be identified. Keywords: Zyryan, word stress, auditory, acoustic, single factor, multiple factor. 1 Introduction to Komi There are two literary Komi languages, Komi-Zyryan (henceforth in this paper Zyryan) and Komi-Permyak. A third Komi dialect, Yaźva, could perhaps be considered a Komi language in its own right. These three Komi forms differ in relation to word-stress. In Komi and its dialects stress can fall on the first syllable, last syllable or on any other. A reliable description of word stress in Komi can be found in V. I. Lytkin (1955). In Zyryan word stress is described as having no fixed position, although there is a tendency for stress to be initially placed, even if this is not obligatory (cf. Lytkin 1970: 248). It must be kept in mind that Lytkin was referring to Komi as spoken in the 1950s, at the latest. In Komi-Permyak the placement of stress varies and in the Yaźva dialect (Eastern Permyak) some vowels attract stress while others do not (Lytkin 1961: 33 34; 1962: 90, 92 94). Lytkin s contemporary Bubrikh (1949), on the other hand, maintained that in Zyryan stress was generally on the first syllable. A more recent description is that of Ye. A. Igushev (1996): as a rule stress falls on the initial syllable, although in some dialects it may alight on -ov, -oo suffixes, on a non-initial syllable in the case of some morphemes or, then again, on suffixes containing the vowels a and ö. In this paper I wish, first of all, to address the question of the placement of word stress in modern-day Zyryan, basing my conclusions on the results of an auditory experiment in which I was assisted by five Zyryan informants. To begin with I shall briefly consider the history of Komi word stress as described in the work of Lytkin and

Estill: On Zyryan word stress 15 on the basis of experiments I myself have conducted, keeping in mind the definitions for the present day mentioned above. 2 The history of Komi word stress While Zyryan word stress is generally on the initial syllable today, the situation in the past has evidently been different. Lytkin has pointed out that the renowned Komi poet M. A. Kuratov even in the 19th century generally marked accent when it did not fall on the initial syllable, which in itself suggests that the position of stress was not then clearly established as being on the first syllable, a practice used by Kuratov in order to indicate the poetic metre (Lytkin 1955: I 60 61). If that were the case, then Zyryan word stress would have been at that time somewhat similar in nature to Erzya which also lends itself to variation in poetic word-stress (A. P. Markov Ocherki. I: 41). A recent collection of some of Kuratov s work (1979) supplies evidence of this practice. Other work by Lytkin (1952) provided even more emphatic evidence of the prevalence of non-initial word stress in earlier centuries. Herein Lytkin considered the nature of Old Permian, the predecessor of modern Komi, including grammar and lexicon, on the basis of extant manuscripts from the 17th and 18th centuries. These manuscripts were marked with an acute accent for word stress. The period of Old Permian commenced in the 1300s, i.e. at the beginning of the historical period, although the evidence in question can probably be most fittingly related to the 17th century. According to calculations made originally by V. I. Lytkin, and confirmed by my own conservative estimates based on the same historical documentation, word stress patterns have changed in such a way that whereas in the 1600s, for which we have historical evidence, word stress was on the first syllable in the case of approximately 70 per cent of words, it is now on the first syllable in the case of approximately 95 per cent of words (see below). 1 Table 1 shows the results of calculations made on the basis of an extant manuscript, the Евгеневский manuscript, which was discovered in the late 18th century and is a Russian transliteration 2 of an Old Permian religious text that was originally written in the Old Permian alphabet. This table provides details concerning the placement of stress on 654 words in the Евгеневский manuscript for which word stress was indicated. The total for unmarked stress is also shown. Table 1. Totals for stressed and unstressed words in the Евгеневский manuscript Total words Stress unmarked Monosyllabic words 3 Stress on σ1 4 Stress on σ1+ 935 281 5 26 283 371 The problem faced in interpreting these figures concerns the number of words that were not marked for stress. These do not seem to have anything in common, or 1 It will be observed that no separation of word and sentential stress has been made in determining the figure for the present day, whereas allowances were made for sentential stress in the historical data. 2 It may be of interest to the reader to know that prior to this period there was a special alphabet for Old Permian, which was later discarded in favour of Cyrillic. Hence, the need for transliteration. 3 This is a fairly rough estimate, because of my difficulty in separating the words in the text. 4 σ = syllable. 5 Includes monosyllabic words.

16 Estill: On Zyryan word stress anything that would single them out from the other words in the text. Furthermore, there were not very many monosyllabic words in the text. We must, then, rely on conjecture. If we assume that unstressed words were omitted because initial syllable stress may have been anticipated as a natural first choice, then calculations of word stress occurrence in the manuscripts become heavily weighted in favour of initial syllable word stress, and calculations of word stress on subsequent syllables is minimised. This makes my resulting estimations very conservative. The figure below shows how the evidence so far can be interpreted on a chart. The curve shows at the very least the changes in word stress that have taken place over, approximately, the last four hundred years. According to this calculation, at least every third word would have been stressed on a syllable other than the first. This situation is visually shown in figure 1. % 100 80 60 40 20 0 1600 1700 1800 1900 2000 year Figure 1. Chart with curve showing a conservative estimate of the possible historical development of word stress on the first syllable in Komi as percentage (y-axis) for the years 1600 2000 (x-axis). 6 3 Zyryan word stress today While word stress in Permyak and the Yaźva dialect shows certain variation despite the prevalence of initial stress, which can well be understood historically in the light of the previous section, in the predominant 7 Komi language, Zyryan, as mentioned above, word stress is generally regarded as being on the first syllable. In order to determine to what extent this is in fact the case today, I carried out an experiment. 3.1 Description of first auditory experiment To discover the extent to which word stress is placed on the first syllable in Zyryan, I 6 The Komi figure has been derived according to my own calculations based on photographs of extant pages of the Евгеневский manuscript published in V.I. Lytkin (1952: Appendix 2). My own original calculation for non-initial syllable stress was 39%, but after tempering this with Lytkin s warning of the presence of sentential stress (I allowed 10% for this) and his observation that 80% of stress was radical and 20% (for which I made a further allowance of 20% which meant that one third would have been radical) suffixal, although no mention was made of the syllable on which radical stress had been placed, I felt that by adding 31% and rounding the final figure to 70 % for initial stress that I would not be accused of exaggerating the prevalence of non-initial word stress in Komi in the 1600s. 7 There are about twice as many Zyryan speakers as there are Permyaks.

Estill: On Zyryan word stress 17 asked a native speaker female Zyryan speaker (SA) to read portions from recently published Zyryan fiction for the purpose of an auditory experiment. My informant is from the village of Kuratovo, Syktyv raion, and the recordings were made when she was 25. She spoke the Middle Sysola dialect. I understand that my informant s pronunciation was quite representative of the Zyryan dialects in general. The reading material was an extract from a story in the book Эжва педымса зонка by V. Timin (2000) describing the adventures of a young boy, and the text was standard. One reason for choosing the text, in addition to its general popularity (I have heard about this from other Zyryans), was the fact that it included very few words that have entered Zyryan from or through Russian. The odd word that did occur was eliminated from the analysis. 8 The recording took place at the Department of Finno-Ugrian Studies of the University of Helsinki in conditions conducive to the production of reasonably good sound quality, at least for the purposes of an auditory experiment. The auditory experiment was accompanied by a somewhat modest acoustic experiment assisted by the same informant, SA. In this case nine short sentences were analysed in order to discover factors which may indicate the extent to which fundamental frequency, intensity and duration were involved in the production of word stress. This was carried out by measuring the acoustic parameters of these correlates. 3.2 Description of second auditory experiment It may not be very convincing to claim that the evidence of one speaker is representative of all, especially in the case of a language in which word stress has not been static. A further auditory experiment was therefore conducted to compare the word stress pattern of several speakers (LC, JG, OM, OO). The dialectal areas of these informants and their profiles are mentioned in section 4.1. Each informant read 274 words onto a CD disk using a Plextalk recorder. The extract used was chosen at random from the abovementioned Эжва педымса зонка by V. Timin. 3.3 Results of auditory experiment 3.3.1 Placement of word stress in first auditory experiment Although SA read several pages of text, for my own analysis it was sufficient to consider only the first 651 words, since the general pattern of initial syllable stress was so predominant. In fact, of those words, only in the case of 35 did stress switch to a syllable other than the first. This meant that 94.6 per cent of words were stressed on the first syllable, and when sentential stress, for which no allowance was made, is taken into consideration it can safely be said that in Zyryan today word stress falls with regularity on the initial syllable. Confirmation of the perceived placement of stress was made by two speakers of languages other than Komi, i.e. English and Finnish, and the experiments were conducted in 2003. This does not, however, mean by definition that were word stress to fall on a syllable other than the first in a neutral environment, that such a pronunciation would be unacceptable, as it would be in e.g. Finnish. Why did 35 words not receive initial stress? When examined more closely the 35 words on which stress did not fall on the first syllable may have been influenced by 8 It is not exactly clear to me, however, what a loan word is and how it should be defined, i.e. how much the form should still resemble that of the language from which it has been derived. For safety s sake recent introductions from Russian (mostly since the October Revolution) were left out.