An Experimental Acoustic Study of Dental and Interdental Nonsibilant Fricatives in the Speech of a Single Speaker * Mark J. Jones

Transcription

1 This document is materially identical to the paper of the same name published in Cambridge Occasional Papers in Linguistics (COPiL) 2, 2005: , but differs in formatting and page numbering. An Experimental Acoustic Study of Dental and Interdental Nonsibilant Fricatives in the Speech of a Single Speaker * Mark J. Jones Department of Linguistics University of Cambridge Dental and interdental non-sibilant fricatives are relatively rare in the languages of the world and are not known to contrast in any language. It is known that a dialect-specific difference appears to operate in English, with American English speakers favouring an interdental realisation and British English speakers a dental one. This study investigates possible motivations for the dialect-inappropriate use of an interdental realisation in the speech of a native British English speaker. An experiment is conducted to see whether a dental realisation of /T/ for this speaker would produce an unsatisfactory [T], e.g. no frication, /s/-like frication etc. It is concluded that the dental realisation of /T/ could function distinctively for this speaker, and that some other factor is responsible for the preferred interdental realisation. A further experiment is conducted to ascertain how frication is generated for dental and interdental fricatives for this speaker. As previously suggested by Shadle (1997), the lips do appear to function as obstacles for the generation of frication. 1 INTRODUCTION The production of a fricative consonant involves a degree of constriction in the vocal tract which will, with an appropriate rate of airflow, generate turbulence. The International Phonetic Alphabet (IPA) provides more unitary symbols for fricative places of articulation than for any other manner of consonant. Among these are the symbols /T/ and /D/, defined as applying to a voiceless dental fricative and a voiced dental fricative respectively (IPA 1999: 177). Fricatives are traditionally cross-classified as sibilants or non-sibilants. The precise definition and use of these terms is disputed, as will be discussed at greater length below. For the present, it is sufficient to note that researchers agree in describing the dental fricatives /T/ and /D/ as non-sibilants. The sounds /T/ and /D/ occur in most varieties of English, but are relatively uncommon cross-linguistically. The UPSID survey of 317 languages (Maddieson 1984) reports that /T/ occurs in 18 languages (5.7% of the sample), and /D/ in 21 (6.6% of the sample), with variants * My thanks to Peter Ladefoged, Michele Loporcaro, and Marija Tabain for their assistance with various parts of the research for this paper. I am solely responsible for any errors which occur by Mark J. Jones Faye Chalcraft & Efthymios Sipetzis (eds.) Cambridge Occasional Papers in Linguistics 2.

2 2 Mark J. Jones of /D/ in a further two languages (23/317 = 7.3%). The dental place of articulation involves a constriction between tip (apex) or blade of the tongue and the upper incisors. However, it is clear that some variation exists in the way individual speakers (and dialects) of English produce /T/ and /D/. British English is said to prefer a dental realisation, whereas in American English an interdental pronunciation is more common (Ladefoged and Maddieson 1996: 143; Cruttenden 2001: ). The interdental realisation involves contact / approximation between the blade of the tongue and the upper incisors and can be described alternatively as laminal dental. More precise descriptions of the articulation of (inter)dental fricatives are provided below. Where required, the interdental articulation can be distinguished from the dental articulation by means of the IPA advanced diacritic, giving the composite symbols [T+], [D+] (Laver 1994, Ladefoged and Maddieson 1996). The linguistic phonetic use of the subscript advanced diacritic will be utilised here. The Extended IPA symbols intended for clinical phonetic use do employ a different diacritic for interdental (and bidental) articulations (IPA 1999: 189, 193). The concentration in what follows is on the voiceless fricative /T/. Interdental and dental non-sibilant fricatives are not known to contrast in any language (Ladefoged and Maddieson 1996: 144). Possibly because of this lack of phonological contrast, the different acoustic characteristics of dental and interdental non-sibilant fricatives have not been addressed in any phonetic study as far as I am aware. As reported above, dialects of English differ as to which realisation of /T/ is preferred. Ladefoged and Maddieson (1996: 143) report that in a survey of 28 native speakers of two broadly defined accent types (British vs. American), an interdental realisation of /T/ was observed for 90% of the American speakers with a dental realisation only evident for the remaining 10%. Among the British English speakers, the pattern was reversed, with 90% of the British English speakers using a dental realisation of /T/, and 10% using the interdental [T+]. Given that a dialect-specific preference is apparent, the motivations for speakers to use a dialect-inappropriate realisation, i.e. individual British English speakers using interdental not dental realisations and American English speakers using dental not interdental realisations, remain to be investigated. This paper attempts to address the question of what motivates a dialectinappropriate articulation of /T/, and to provide some more phonetic data on the acoustic differences between voiceless dental and interdental non-sibilant fricatives. 2 THE PHONETICS OF DENTAL AND INTERDENTAL NON-SIBILANT FRICATIVES 2.1 Cross-linguistic variation in production As remarked above, no language appears to contrast dental and interdental non-sibilant fricatives. English exhibits both dental and interdental realisations of /T/. Spanish /T/ is described by Navarro Tomas (1968), and Martínez-Celdrán, Fernández-Planas, and Carrera- Sabaté (2003) as interdental. Welsh is regarded by G. E. Jones (1984: 46) as having dental non-sibilant fricatives /T/ and /D/. Icelandic (Pétursson 1971) and Greek too are regarded as having dental realisations (Arvaniti 1999a, b). In the Greek case, it may be here that the IPA label is being used in a broad sense without closer phonetic inspection of speaker articulations. An informal survey of four native speakers of Greek living in Cambridge finds

3 Dental and Interdental Fricatives 3 that the interdental articulation is common to all 1. Given that no phonological contrast between dental and interdental variants is observed, the broad or phonological use of the IPA label for /T/ and /D/ is not unlikely, and has definitely occurred in studies on English (see below). It is not clear to what extent patterns of dialectal and individual variation similar to those reported in English occur in other languages which have /T/ and /D/. Some indication that variation might occur comes from data on Icelandic. Although Icelandic /T/ is described by Pétursson (1971) as having a dental realisation, the data come from a single speaker in addition to introspection on Pétursson s own speech habits. Another source of data on Icelandic presents palatograms showing interdental realisations of /T/ and /D/ (Einarsson 1945: 13). Pétursson (1971) also reports a difference in articulation between /T/ and /D/, a pattern which remains to be confirmed for other languages or explained, though it may be due to the different production requirements for voicing in /D/. The mismatch between descriptions of Greek /T/ as dental in Arvaniti (1999a, b) and the interdental result of the informal survey of four Greek speakers conducted in Cambridge may testify to similar variation in Greek. Further data is urgently required. 2.2 The interdental-dental continuum Cruttenden (2001) describes the production of dental /T/ and /D/ in the variety of British English called Received Pronunciation. He states that: (1) the tip and rims of the tongue make a light contact with the edge and inner surface of the upper incisors and a firmer contact with the upper side teeth, so that the air escaping between the forward surface of the tongue and the incisors causes friction (such friction often being very weak in the case of /D/). (Cruttenden 2001: ) As for the interdental realisation, Laver (1994: 250) states that tongue tip protrudes between the upper and lower teeth in the interdental realisation of /T/ in Californian English. The issue of whether to classify an articulation as dental or interdental seems fairly clear-cut in previous descriptions, but is complicated by observations that very many dental realisations can involve some tongue tip protrusion. It is well-known that dental realisations usually involve some contact of the tongue blade with the upper incisors and alveolar ridge for speakers with relatively normal dentition, and therefore for many speakers a dental realisation may involve some protrusion of the tongue tip beyond the upper incisors. G. E. Jones (1984: 46) remarks that the dental fricatives /T/ and /D/ in Welsh have lamino-dental contact with the tip of the tongue slightly protruding between the teeth. Ladefoged and Maddieson (1996: 251) comment on interdental realisations of click consonants in Sandawe and Hadza which are usually described as dental clicks, and refer also to Doke (1925) who offered similar reports on interdental realisations of clicks in the!xu) language. Ladefoged and Maddieson (1996: 44) also report interdental realisations of dental /n5/ for some speakers of the Dravidian language Malayalam. 1 My thanks to Marios Mavrogiorgos for his assistance in conducting this informal survey.

4 4 Mark J. Jones The observation of interdental realisations of /n5/ in Malayalam leads Ladefoged and Maddieson to comment on the continuum between alveolar, dental and even linguo-labial places of articulation. As such, it may seem unwise to attempt to divide articulations from various speakers and languages into two distinct groups. However, a truly laminal interdental articulation, i.e. formed with contact between the upper incisors and the tongue blade (or sides of the tongue blade in the case of fricatives) results in substantial protrusion of the tongue tip (see figure 1 in section 4 below). The profile of the tongue behind the constriction may also differ, though this is impossible to observe without instrumental techniques. The division between interdental and dental realisations of /T/ in the literature on English seems to be one that has caused very few classificatory problems, suggesting that articulations of the fricative /T/ tend to fall into fairly distinct regions within the continuum which undoubtedly exists. Cross-linguistic data appear to bear this out: interdental realisations of other more usually dental sounds, such as those listed above for dental /n5/ and the clicks, seem to be rare, and worthy of particular comment. Thus while it is apparent that interdental realisations of an otherwise dental speech sound can occur to varying degrees, the extent of tongue-tip protrusion and possibly also the nature of the laminar contact seem to be particularly common and extreme with some realisations of /T/. Interdental realisations of non-fricatives seem to be generally rare phonetically and do not seem to contrast with any dental realisations of the same manner of consonant. Interdental stops are not reported to occur linguistically in any language, though there is some suggestion that they might occur in American English due to coarticulation to a neighbouring interdental /T/ (Lavoie 2003). Interdental /n/ and /l/ also seem to occur primarily due to lingual coarticulation to a following interdental /T/ in Spanish e.g. in encía [en+»t+ia] gum, and alzar [al+»t+ar] to raise (Martínez-Celdrán et al. 2003: 258; symbols adapted), and interdental /l/ may also occur due to coarticulation to interdental /T/ in various varieties of English. Interdental /l/ may also occur without any obvious coarticulatory origin in some varieties of English, though little is known of the motivation for or geographical and sociolinguistic distribution of these interdental realisations of /l/ (Jones 2004a, b). Other non-coarticulatory interdental forms are rare. Loporcaro and Mancuso (1998) report interdental /l/ as the historical development of non-geminate /l/ in some southern Italian dialects. The acceptance of a clear division between dental and interdental realisations of /T/ in the face of a potential continuum of articulations rests on the historical tradition of descriptions of English and other languages, the fact that interdental realisations of otherwise dental sounds are rare enough to be remarked upon specifically, and cross-linguistic patterns of contrast. Articulatory preferences for realisations at the ends of the dental-interdental continuum is a subject which requires more research. Until more phonetic data are forthcoming, it seems not unlikely that there is an affinity for /T/ to involve an extreme interdental realisation in some languages or dialects or for some speakers, a possibility which is in conformity with the wider range of places of articulation seen for fricatives compared to other consonantal manners. 2.3 Acoustic and perceptual studies of interdental and dental non-sibilants Previous acoustic studies of /T/, largely from American and less commonly British or Australian English, fail to distinguish explicitly between the two possible realisations of /T/,

5 Dental and Interdental Fricatives 5 dental or interdental, among their subjects. Again this lack or phonetic acuity may be due to the observation that no language is known to contrast dental and interdental fricatives. Tabain (2001) and Jongman et al. (2000: 1256) employ a broad use of the term dental to refer to the articulation of /T/. The production of /T/ is also described by Jongman et al. (2000: 1252) as (inter)dental. Tabain (p.c.) herself uses an interdental realisation, but the articulatory preferences of the other subjects in Tabain (2001) were not specifically noted. Consequently, it is not known how the dental and interdental realisations of /T/ differ acoustically, and existing data are unhelpful on this point. Given that most data come from speakers of American English, and assuming that the survey of Ladefoged and Maddieson (1996) is representative, it might be assumed that acoustic descriptions so far are more likely to report on interdental realisations. Strevens (1960) is an exception in that the subjects are speakers of British English, and may contain acoustic data on dental realisations of /T/. Previous acoustic studies agree that the amplitude of frication in /T/ tends to be fairly low relative to /s/ and /S/, but is more comparable with the amplitude of /f/ (Strevens 1960; Tabain 1998). The spectral shape of /T/ is very flat, with no major peaks, and the extent of the fricative energy covers a wide frequency range, typically 1000 Hz 10,000 + Hz. The spectral characteristics of /T/ and /f/ are quite similar (Strevens 1960; Jongman et al. 2000: 1253). This spectral similarity may underlie reportedly high rates of confusability between the two in general speech and also under laboratory conditions. Harris (1958) showed that speakers of American English were unable to distinguish /T/ and /f/ on the basis of the frication alone. Successful perceptual contrast appears to be mainly carried out on the basis of formant transitions. This is also seen for the /T/ ~ /f/ contrast for male speakers of Toda, whereas the fricative contrasts of female speakers could be distinguished in terms of duration (Gordon et al. 2002: 170). More cross-linguistic research on the production and perception of /T/ and /f/ is required. 2.4 (Inter)Dentals as non-sibilants: defining sibilance The realisation of /T/ in English is universally described as being non-sibilant. The sibilant ~ non-sibilant distinction among fricatives is a longstanding one in English. The fricatives /s/ and /S/ are described as sibilant, and /T/ and /f/ are regarded as being non-sibilants. The phonetic property of sibilance has been linked with a phonological feature [strident]; at least, the two terms are currently synonymous for most researchers (Kenstowicz 1994; Laver 1994: 260; Ladefoged and Maddieson 1996: 138; Stevens 1998). This is not the case for Shadle (1990, 1997), who maintains an undefined distinction between sibilant and strident. Sibilant is associated directly with the term obstacle fricative by Ladefoged and Maddieson (1996: 138). The term obstacle fricative refers to the generation of turbulence and hence noise at an obstacle downstream of the constriction (Shadle 1990). Stevens (1998) uses the same aerodynamic definition for a phonological feature [strident]. He states that (2) The feature [strident]... indicates whether an obstacle is placed in the airway downstream from the constriction, to enhance the acoustic excitation of the front cavity [of the vocal tract] by turbulence noise. Thus the consonants /s, S/ are [+strident], whereas /T, D/ are [ strident]. (Stevens 1998: 249)

6 6 Mark J. Jones Traditionally strident has a more acoustic-auditory basis referring to amplitude or intensity and pitch (Laver 1994: 260). Shadle (1990: 206) appears to endorse the association of obstacle fricatives and a quality of sibilance, but takes issue with drawing a simple parallel between non-sibilants and non-obstacle fricatives. She notes that another type of fricative is the wall-source fricative in which frication noise is generated by turbulence created against a surface running parallel to the direction of airflow, and that in effect, a continuum is likely between wall and obstacle fricatives (Shadle 1990: 206). A dichotomous phonological or phonetic classification as sibilant and non-sibilant (or terminological equivalents) must be based on other aspects of the fricative production as well. There is therefore some debate about whether or not the term non-sibilant should be applied on the basis of a characterisation of a fricative as an nonobstacle fricative. As Shadle herself comments: (3) The lack of agreement regarding which feature is more useful [strident or sibilant], or indeed which fricatives ought to be in which group, reflects our poor understanding of the acoustic mechanism of fricatives. (Shadle 1990: 187). Shadle (1997) comments further that the non-sibilant fricatives of English may be neither wall nor obstacle fricatives, but that they may constitute a special case: (4) The weak front fricatives [f, T] should also possibly be grouped in [the non-obstacle wall fricative] category, since noise is clearly generated along the lips... The wall does not continue very far, however, and so it may be that these sounds should be considered as a third category. (Shadle 1997: 47). The latter comment also indicates the lack of agreement in the literature about the precise mechanism for the production of the turbulence in /T/. Ladefoged and Maddieson (1996: 144) state that both the American and British varieties of T and D are non-sibilant fricatives with the turbulence being produced at the interdental or dental constriction. The classification of /T/ as a non-sibilant is clearly based on its non-obstacle status. Cruttenden (2001: ) describes the production of frication for /T/ and /D/ in RP as follows: the air escaping between the forward surface of the tongue and the incisors causes friction (such friction often being very weak in the case of /D/). In Californian English Laver (1994: 250) states that the protrusion of the tongue tip between the upper and lower teeth in the interdental realisation of /T/ allow[s] the airflow between the tongue surface and the upper teeth to become turbulent. There is general agreement that the production of turbulence for /T/ and /D/ (whether dental or interdental) is at the point of constriction in English, and this claim is echoed for Welsh by G. E. Jones (1984: 46): [in /T/ and /D/] the air escapes with friction between the blade of the tongue and the upper incisors. For most researchers, therefore, the frication is generated at the point of constriction rather than at a downstream obstacle, and this directly underlies the basis for describing /T/ as a non-sibilant in Ladefoged and Maddieson (1996: 138), and a phonological specification as [-strident] in accordance with Stevens (1998: 249). As Shadle s comments cited in (3) above make clear, it is her opinion that the lips are involved in the production of frication in /T/, and constitute some form of obstacle. The wider issue of how to define sibilance, and whether or

7 Dental and Interdental Fricatives 7 not this can be equated with stridency will not be addressed in detail here, except to note that some dispute exists in the literature. The current paper will have as one of its research aims an experimental assessment of the role of the lips in the generation of frication for dental and interdental non-sibilant fricatives. 3 POSSIBLE CAUSES OF REALISATIONAL VARIATION The results of the survey of British and American English articulatory variation in the production of /T/ reported in Ladefoged and Maddieson (1996) show strong dialect-specific preferences for interdental (90% American English) or dental (90% British English) realisations. Previous phonetic studies of interspeaker variation in the production of other speech sounds have shown that the idea of a uniform production strategy in terms of place of articulation for all speakers of a single language variety is false. Dart (1991) showed that speakers of French and English did not pattern as expected in their production of coronal stops. French /t/ is usually described as dental and English /t/ as alveolar. However, in Dart s sample, a large minority of speakers of both languages, almost 50% in fact, used an articulation for /t/ which would have been classified as a different place of articulation from that expected. Given these figures in variation in stop production and the non-attestation of contrasting dental and interdental non-sibilant fricatives in any language, it might be expected that variation between speakers in the production of /T/ would occur. However, on the basis of Dart s results and the general observation that dental and interdental /T/ do not contrast, it might also be expected that variation would settle around the 50% mark for each variant. The observed use of an articulation which might be described as non-appropriate for the dialect in question, i.e. dental for an American English speaker or interdental for a British English speaker, among such a small minority of speakers (10%) is therefore surprising. Two possible sources of variation are addressed below: physical causes, and linguistic causes. 3.1 Physical causes of variation As Ladefoged and Maddieson (1996: 137) point out, a great deal of precision is required in producing a fricative: [I]n a fricative a variation of one millimeter in the position of the target for the crucial part of the vocal tract makes a great deal of difference. It seems likely, therefore, that variation in the shape or morphology of the vocal tract may have a large influence on individual production strategies for fricatives. Ladefoged and Maddieson (1996: 146) refer to a study on the influence of differences in the patterning of the ridges (rugae) of the hard palate on the production of /s/, but in general the influence of vocal tract morphology on articulatory preferences in non-disturbed speech has hardly been studied (Beck 1997: 257). It seems possible that the preference for interdental over dental realisations could be due to variation in some aspect of vocal tract morphology or dentition. Jespersen ( , cited in Ladefoged and Maddieson 1996: 144), considered that the variation may be due to the size of gaps between the teeth (interstices), with interdental realisations preferred by speakers who had large gaps. Strevens (1960) also notes the influence of dentition on the production of /T/:

8 8 Mark J. Jones (5) /T/ [in British English] is produced with a narrow slit between the bottom of the upper teeth and the surface of the tongue; the configuration of the teeth affects the quality of the sound produced. (Strevens 1960: 34) There could therefore be some anatomical cause for variation between interdental and dental preferences for the articulation of /T/. Some speakers of British English may have vocal tract morphology which makes a successful production of /T/ using a dental articulation impossible. For these speakers (10% according to the survey of Ladefoged and Maddieson 1996), an attempt to produce a dental realisation of /T/ may fail to generate frication, or, as may be hypothesised in the basis of Strevens comment in (2) above, the frication noise so produced may be unrecognisable as /T/. However, it seems unlikely that two populations, speakers of American English and speakers of British English, should differ so much in their articulatory preferences in such a complementary way when numerous unidentified aspects of widespread individual variation in vocal tract morphology may play a role. 3.2 Linguistic causes of variation It is also possible that linguistic causes of variation exist. Ladefoged and Maddieson (1996: 20) comment that some speakers of the Dravidian language Malayalam they observed used interdental realisations of canonically dental /n5/, a practice not extended to the dental /t5/. This they ascribed to the more salient bursts of the plosives and suggested that (6) [t]hose speakers of Malayalam who have interdental nasals might thus increase the difference [between phonologically dental and alveolar nasals] by producing more distinct formant transitions as a result of the interdental articulation. Citing work carried out by Dart (1991, 1993), they comment further that apical alveolar stops have a higher F2 and lower F3 and F4 than laminal dental stops (Ladefoged and Maddieson 1996: 25). It might therefore be expected that laminal interdental fricatives show a similar pattern relative to apical dental fricatives. In English, however, no obvious contrast could be enhanced by this mechanism, as there is no contrasting non-sibilant fricative at an alveolar place of articulation. There may also be a link between an interdental realisation of /T/ and a dental realisation of /t/, if, as Lavoie (2003) claims, (inter)dentals are prone to being realised as stops in casual and connected speech. A speaker who has a dental /t/ may therefore use an interdental /T/ to distinguish the canonical realisation of the former from the stopped casual speech realisation of the latter. This hypothesis is not supported by the 50% non-dental figure for American English /t/ recorded by Dart (1991), though the subject of the experiments reported here, the author, does have a dental realisation of /t/ and an interdental realisation of /T/, as it would appear does the subject (if it was a single subject) who provided the Icelandic palatographic data for Einarsson (1945). These issues are noted but are not explored further here. In order to investigate the motivation for a dialect-inappropriate realisation of /T/ further, an experiment will be conducted using the author as a subject in which the dialectappropriate realisation (dental), dispreferred by this speaker, will be compared with the

9 Dental and Interdental Fricatives 9 subject s preferred interdental realisation of /T/ and also realisations of /s/ and /f/. Using data from a single subject will control for interspeaker variation in the acoustic correlates of fricatives. In this way, any aspect of the dispreferred dialect-appropriate dental realisation of /T/ may be identified. The role of the lips in the generation of frication in the production of /T/ will also be investigated for this subject in a further experiment. 4 EXPERIMENT 1: AN ACOUSTIC STUDY OF DENTAL AND INTERDENTAL /T/ 4.1 Aspects of the subject s vocal tract morphology and interdental realisation The subject was the author, a 32 year old, male, phonetically-trained native speaker of British English. A detailed articulatory study of the subject such as that which might be afforded by MRI or x-ray has not been carried out. The number of potential factors arising from variation in vocal tract morphology is immense, e.g. general dentition, more specific aspects of dentition, curvature of upper jaw (maxillary arch), curvature of lower jaw (mandibular arch), slope and height of palate, size and relative location of alveolar ridge, palate width, point of attachment of the tongue to the floor of the mouth, number and pattern of rugae etc., etc. For this reason, the experiment described below concentrates on the acoustic output of the dispreferred dialect-appropriate realisation of dental /T/ in comparison with other fricatives. However, a number of general comments on the subject s dentition and articulatory preferences can be made. Beck (1997: 285) states that around 50% of the population may have some mild malocclusion, i.e. an abnormal relationship between adjacent teeth or between antagonist teeth in opposing jaws. One class of malocclusion, angle class II, described as being when a subject has the mandibular arch (lower jaw) posterior to the maxillary arch (upper jaw), accounting for around 38% of all malocclusions (Beck 1997: 288), has been linked to interdental realisations of some speech sounds, notably /s/. The subject has minor malocclusion of the class I type: some teeth are misaligned relative to their neighbours, but the relationship between the upper jaw (maxilla) and lower jaw (mandible) seems normal and no lisping of /s/ occurs. Interestingly, given Jespersen s speculation about the role of interstices (gaps between the teeth) in the production of frication for interdental /T/, the subject has no noticeable gaps, showing instead some overcrowding of teeth in the mandible. The production of an interdental stop is possible for this subject, and in the case of a voiced interdental stop [d+], closure is sufficiently complete to induce passive devoicing due to the equalisation of intraoral and subglottal pressure during stop closure (Ohala 1983). Figure 1 shows 9 frames covering 270 ms of the production of interdental [T+] by the subject. The word was read in isolation from a list of rhyming words (sigh, lie, tie, thigh). The articulation takes approximately 270 ms to complete with maximum constriction achieved in frame 6 (rightmost frame in second row down ms). The interdental nature of the articulation is evident around 5 mm of the tongue tip protrudes (as judged separately). The upper central pair of incisors protrude below the level of the incisors either side, with the edge of one flanking incisor just visible on the left in frames 8 and 9. There is no gap running between the central upper incisors what appears as such in the frames is in fact a small notch in one tooth.

10 10 Mark J. Jones Figure 1: Webcam images showing the articulation of interdental /T/ in thigh by the author (sample rate 30 Hz time course of articulation runs from left to right along each row from frame 1 to frame 9) The extent of variation within the speech of a single speaker is also unknown. In my own speech, instances of /T/ occurring in intervocalic, final pre-pausal, and utterance-initial positions are all interdental. On the whole, interdental realisations dominate my own production preferences, though dental realisations do seem to be possible, and even perhaps more common in unemphatic speech, when following /n/ or non-vocalised /l/, and also when following /s/, /z/, /d/ or a /t/ (even if the latter is realised as [/]). Whether a word or morpheme boundary intervenes or not appears to be irrelevant. In these cases, the fricative nature of the /T/ is much diminished, and a dental stop [t5] or affricate [t5t] appears to be the impressionistic result (cf. Lavoie 2003). These comments on articulatory variation in a single individual are based on observations of speech which naturally run the risk of being highly susceptible to the observer s paradox, and more objective sources of data are urgently required. 4.2 Data elicitation, recording and analysis Six repetitions of 4 isolated fricative-initial words were elicited. In all cases the fricative for analysis was followed by the diphthong /ai/. The words used in the analysis were sigh, thigh and fight. The word thigh was elicited twice in sequence: first with an interdental fricative (the preferred realisation of the subject) and then with a dental fricative (a realisation which occurred sporadically in the subject s speech due to coarticulation to a preceding coronal) in each series of repetitions. A total of twenty-four test words were elicited for analysis (6 repetitions x 4 fricatives).

11 Dental and Interdental Fricatives 11 The data were recorded using a good quality AKG microphone and Damien preamplifier directly as WAV files onto a Creative Nomad Jukebox 3 portable hard-drive in the sound treated booth in the phonetics laboratory at the University of Cambridge. The sampling rate was 22,000 Hz. The data were then transferred to a Toshiba laptop computer running Microsoft Xp for low-pass filtering at 9,700 Hz and analysis. The analysis used Sensimetrics Speech Station 2 speech analysis software except for the calculation of Centre of Gravity (COG), which was carried out using Praat (see below). The fricatives were analysed in terms of the onset of frication in the frequency range (low frequency cut-off) as determined from a wideband spectrogram. Low frequency cut-off was measured by eye using the cursor in the spectrogram window. The amplitude and frequency of the major spectral peak were measured from an average FFT spectrum (256 sample window size 64 sample inter-fft interval) calculated from a 50 ms slice around the midpoint of each fricative. In order to normalise for possible variations in peak spectral amplitude between repetitions, the amplitude of F2 in the following vowel was also measured from a similar average spectrum calculated from a 50 ms slice at vowel onset. A proportion of the amplitude of the spectral peak to F2 amplitude was calculated (spectral peak amplitude/f2 amplitude). This is not the same amplitude normalisation procedure used in e.g. Jongman et al. (2000) which varied the particular vowel formant against which the fricative peak amplitude was normalised, but allows all fricatives and all vowels to be compared in the same way. The range of generally uninterrupted frequency peaks falling within 12 db of the peak amplitude was also determined from the average spectrum of the fricative. All frequency measures were recorded in Hertz and were converted also into Bark values. The COG measures were calculated automatically using the COG command in the Praat speech analysis software package. The measure of COG takes the overall amplitude and frequency characteristics of the fricative into account (see Ladefoged 2003: ; Gordon et al. 2002). 4.3 Results of experiment Amplitude values The graph shown below in figure 2 shows the average amplitude values normalised against F2 of the following vowel. The absolute values for the vowel amplitude were very comparable across repetitions (overall average 35.7 db, s.d. 1.5, n = 24). The non-sibilants /f/ and [T±] and [T] show a much lower average amplitude than /s/, for which the amplitude of the spectral peak exceeded that of F2 of the vowel. In the case of /f/ and [T±], the amplitude of the spectral peak was around 50% of that of F2 of the following vowel, but for [T], the amplitude was still lower at around 30%.

12 12 Mark J. Jones Fig 2: Graph showing the average amplitudes (db) of the spectral peak of each fricative as a proportion of the average amplitude of F2 of the following vowel. The average amplitude of the spectral peak of the sibilant /s/ exceeded that of F2 of the following vowel. The average amplitudes of /f/ and interdental /T/ (labelled idth) were also comparable, as expected. The average amplitude of the spectral peak of dental /T/ (labelled dth) was lowest in relation to the F2 of the following vowel. Average amplitude proportions Low-frequency cut-off The values of the low frequency cut-off show a clear pattern of sibilant vs. non-sibilants for /s/ vs. /f/ and [T±], but the values for [T] are the highest of any fricative in the study (see table 1 below). Table 1: Table showing the average values for low frequency cut-off (Hz and Bark). /s/ [T±] [T] /f/ average cut-off (Hz) s.d average cut-off (Bark) s.d n Figure 3 below shows average spectra for single tokens of each fricative. Note that the amplitude scale for /s/ (vertical scale) differs from that of the other spectra in that it runs from around 20 db to over 50 db. The other spectra run from around 0 db to around 40 db. A similarly-scaled /s/ spectrum would have omitted all spectral detail. The horizontal black line in each spectrum shows the level of 30 db, allowing amplitude comparison between /s/ and the other fricative spectra to be made.

13 Dental and Interdental Fricatives 13 Figure 3: Average FFT spectra (256 window size, 64 sample inter-fft interval) of 50 ms of the midpoint of individual tokens of each fricative. As described in the text, the amplitude values are comparable for interdental [T±], dental [T], and [f], but not for [s]. A horizontal black line marks the 30 db level for comparison between [s] and the others. The essentially flat nature of the spectra of the non-sibilants is apparent compared with sibilant [s], though dental [T] has a more distinct spectral shape. The vertical line indicates the location of the major spectral peak. For [f], two peaks come into contention but only the lowest frequency peak is marked. For the purposes of calculation of average peak frequencies, the peak which most resembled previous peak frequency values was selected (see text). [s] interdental [T±] dental [T] [f] Peak frequency In all but two cases, once for /f/ and once for dental [T], a single major amplitude peak could be identified in the spectrum. In these two aberrant cases, two peaks widely separated in the spectrum had the same amplitude. For the purpose of calculating the average values presented in table 2 below, the peak frequency which most approximated previously identified peaks was included.

14 14 Mark J. Jones Table 2: Table showing the average values for peak frequency (Hz and Bark). /s/ [T±] [T] /f/ average peak (Hz) s.d average peak (Bark) s.d n A single outlier of 5725 Hz has been omitted from the average calculation for [T±]. As can be seen from table 2, /s/ has a relatively low frequency peak, though it occurs at a much higher amplitude. For the non-sibilants, especially /f/ and [T±], peaks in the upper reaches of the frequency scale dominate the spectrum. The average peak frequencies are also illustrated in the graph for range in figure 4 below Frequency range In general it was easy to identify the extent of peaks falling within 12 db of the amplitude of the major spectral peak. In the case of /s/, a region of relatively low amplitude separated the region around the spectral peak from a region of higher amplitude within the 12 db range in the high frequencies (7050 Hz 9700 Hz). In two cases [T±] showed a low frequency peak within the range at around 2600 Hz. A low frequency peak was common for [T], occurring between 2366 Hz and 3458 Hz in all cases but one. The average ranges (Hz values) and the location of the average spectral peak frequency within that range is shown in figure 4.

15 Dental and Interdental Fricatives 15 Figure 4: Graph showing the average frequency range (vertical line) for each fricative together with the location within that range of the average frequency of the major spectral peak in each case (horizontal bar). The category labels s and f on the x-axis are selfexplanatory; idth stands for the interdental fricative, and dth stands for the dental fricative. Average frequency (Hz) Fricative The graph in figure 4 shows that /s/ on average has the shortest range (1943 Hz) with a spectral peak at 4676 Hz and that dental /T/ has the next widest range (5494 Hz) and a peak at 6383 Hz. In terms of both range and peak location, dental /T/ differs from interdental /T/. Interdental [T±] has a wider range (6914 Hz) and a higher peak at 8014 Hz. With a range of 6765 Hz and a peak at 9295 Hz, /f/ closely resembles interdental /T/. Note that the lower value of the range is not the same thing as the value for low frequency cut-off (section 5.2.2) Centre-of-Gravity (COG) The figures for COG are presented in table 3. below. Table 3: Table showing the average values for Centre-of-Gravity (COG) in Hz and Bark. /s/ [T±] [T] /f/ average COG (Hz) s.d average COG (Bark) s.d n

16 16 Mark J. Jones Once again, /f/ and interdental /T/ are comparable in terms of absolute values and dispersion of the results. The other fricatives, /s/ and [T], show relatively little dispersion of values, with /s/ having COG averages closer to /f/ and interdental /T/ than [T]. 4.4 Summary of the results for Experiment 1 The results of experiment 1 show that the subject s non-preferred dental fricative [T] is distinct from other fricatives /s/ and /f/ in terms of amplitude and spectral parameters. It does not appear to be the case, therefore, that the dental realisation is unsuitable in terms of being confused with another contrastive fricative. Interestingly, the subject s preferred interdental realisation of [T+] shows many more points of similarity with that of /f/, and the confusability of /f/ and (possibly) interdental /T/ has been noted in previous research (see above). The amplitude of dental /T/ is lower than all the other fricatives, and it may be that this is a possible factor in the subject s preference for interdental [T+], even if the latter shows more spectral similarity to /f/. Other possible linguistic factors are discussed in section 6 below. 5 EXPERIMENT 2: THE GENERATION OF FRICATION FOR DENTAL AND INTERDENTAL /T/ 5.1 Data elicitation, recording and analysis The subject was the same 32 year old phonetically trained native speaker of British English who participated in experiment 1 (the author). The subject was recorded producing sustained voiceless dental and interdental fricatives. At the sound of a self-administered audible cue (a click of the fingers), the subject manually retracted the upper lip by placing the hands around the head, fingers around the back of the skull, and drawing the upper lip backwards and up with the thumbs. This position was maintained during continued production of the fricative in order to determine the effect of upper lip retraction. The click is detectable in sound recordings, and allow a comparison of fricative production before and after lip retraction. This procedure was repeated separately with both dental and interdental fricatives. A second part of the experiment followed the same pattern with the subject producing a sustained voiceless dental or interdental fricative, but after the audible cue, the lower lip was manually retracted by placing the index fingers at either side of the mouth and drawing the lower lip down. Again this position was maintained during continued production of the fricative in order to determine the effect of lower lip retraction. As the experiment involved a real vocal tract and not a model, it was not possible to vary the position of one lip completely independently of the other lip. Some distortion of the non-retracted lip was inevitable during retraction. Distortion of the non-retracted lip may have had some influence on the resulting patterns of frication, as discussed below. The data were recorded on the same occasion and analysed using the same equipment as described above for experiment 1. Visual analysis of the data from wideband spectrograms (128 samples) was carried out. The analysis observed in quite gross terms whether or not

17 Dental and Interdental Fricatives 17 retraction of the upper or lower lip affected the production of frication. The results are presented below. 6.2 Results for experiment Dental fricative upper lip retraction The spectrogram in figure 5 below shows the consequences of manually retracting the upper lip while producing a sustained voiceless dental non-sibilant fricative. Fig. 5. Spectrogram showing the effect of upper lip retraction on the production of a voiceless dental non-sibilant fricative. The transient at around 11.3 on the time axis (marked by arrow) is the click cue for the onset of lip retraction. The spectrogram in figure 5 shows the sustained fricative with no lip retraction until the audible cue (the click) which has been marked by an arrow on the spectrogram. Some delay (1017 ms) occurs between the cue and the acoustic consequences of upper lip retraction. Prior to lip retraction the low frequency cut-off of frication measured from a 256 point average spectrum of 100 ms of sustained frication occurs at 5248 Hz. After lip retraction, from around 12.3 on the horizontal time scale, the low frequency cut-off occurs at 7974 Hz Dental fricative lower lip retraction The spectrogram in figure 6 below illustrates manual retraction of the lower lip during production of a sustained voiceless dental non-sibilant fricative.

18 18 Mark J. Jones Fig. 6. Spectrogram showing the effect of lower lip retraction on the production of a voiceless dental non-sibilant fricative. The transient at around 25.5 on the time axis (marked by arrow) is the click cue for the onset of lip retraction. The spectrogram in figure 6 again shows sustained frication (low amplitude) with lip retraction taking place around 753 ms after the click cue (marked by the arrow). A marked change is seen in the gross spectral characteristics of the fricative, with a complete loss of frication after Interdental fricative upper lip retraction The consequences of manually retracting the upper lip while producing a sustained voiceless interdental non-sibilant fricative are shown in figure 7 below. Fig. 7. Spectrogram showing the effect of upper lip retraction on the production of a voiceless interdental non-sibilant fricative. The transient at around 29.2 on the time axis (marked by arrow) is the click cue for the onset of lip retraction. Following a period of sustained frication, a delay of 824 ms can be seen between the click cue (marked again by the arrow) and spectral changes due to the retraction of the upper lip. The spectrogram in figure 7 shows that a change is seen in the gross spectral characteristics of the fricative from around 30 on the horizontal (time) scale. The lower cut-off boundary of

19 Dental and Interdental Fricatives 19 frication shifts upwards from around 2000 Hz to around 5000 Hz and a diminution in amplitude also occurs Interdental fricative lower lip retraction The spectrogram in figure 8 illustrates the consequences of manually retracting the lower lip during production of a sustained voiceless interdental non-sibilant fricative. Fig. 8. Spectrogram showing the effect of lower lip retraction on the production of a voiceless interdental non-sibilant fricative. The transient at around 2.5 on the time axis (marked by the arrow) is the click cue for the onset of lip retraction. Again, the audible cue (the click) marked on the spectrogram in figure 8 shows the point at which lip retraction is initiated. No obvious and consistent change is seen in the gross spectral characteristics of the fricative as a result of lower lip retraction in the production of the interdental realisation of /T/. 5.3 Discussion of the results of Experiment 2 The results of Experiment 2 show clearly that for this subject, generation of frication for the voiceless interdental non-sibilant fricative is mainly at the upper lip. Manual retraction of the lower lip has no major spectral consequences for the interdental production of frication (figure 8), but retraction of the upper lip results in a marked spectral change (figure 7). Even with retraction of the upper lip, some frication continues to be generated with the interdental articulation. It is most likely that this is generated at the point of interdental constriction (tongue blade lower edge of upper incisors), though it may also be that the jet of air strikes some part of the distorted labial opening. For the dental fricative, upper lip retraction has some consequences for the production of frication, notably an upward shift in the low frequency cut-off from 5248 Hz to 7974 Hz (figure 5), but manual retraction of the lower lip results in a complete loss of frication for this speaker (figure 6). The lower lip appears to be more important for the generation of frication in the dental fricative.

20 20 Mark J. Jones 6 DISCUSSION The main motivation for this investigation of relatively rarely studied fricatives was to ascertain whether a British English speaker who used the dialect-inappropriate interdental fricative realisation of /T/ instead of the dialect-appropriate dental realisation of /T/ did so because the dental realisation was somehow unsuitable. It was hypothesised that this unsuitability might be due to dental [T] having acoustic characteristics which were too similar to another fricative, e.g. /s/. Presumably these acoustic characteristics would be due to some (unassessed, and possibly unassessable) aspect of the subject s vocal tract morphology. This study has shown that for this speaker, data from realisations of /s/, /f/ and interdental /T/ accord well with results reported elsewhere for these English fricatives (e.g. Jongman et al. 2000), assuming that the /T/ of studies of American speech has a 9 in 10 chance of being interdental. Little data is available for a definitely dental realisation of /T/ in English, perhaps only Strevens (1960). Similarities in the relative values of Centre-of-Gravity (COG) for the relevant fricatives are also apparent from the data on the dental /T/ of Toda in Gordon et al. (2002). The acoustic similarity of /f/ and interdental /T/ is exhibited in all acoustic parameters measured here. Unlike /s/, these fricatives have relatively low amplitude and show a flat spectral profile with perhaps some higher frequency dominance. The range of frequencies within 12 db of the peak amplitude is extensive for /f/ and interdental /T/. In contrast, dental [T] shows a number of characteristics which set it apart from /s/, /f/ and interdental /T/. It has the lowest relative amplitude of all the fricatives examined in this study, it has a distinct as opposed to a flat spectral profile, and a COG which testifies to the dominance of the high frequency components. In terms of spectral shape, dental /T/ is more like /s/ than the others, but it shows a higher frequency cut-off than all other fricatives, and has a medium range. It is apparent that dental /T/ is spectrally distinct from all other fricatives for this speaker, including /s/. Not only is dental /T/ distinct from /s/, it is also markedly less similar to /f/ than the speaker s preferred interdental [T±] realisation. The hypothesis that the dental realisation is dispreferred on the grounds of spectral similarity to (and therefore potential perceptual confusability with) another fricative must be rejected. It may be that a dental realisation of /T/ is not in use by this speaker because the frication generated in this way falls too far below some acceptable amplitude threshold. As detailed above, [T] has a relative amplitude only 33% of that of F2 in the following vowel. Another explanation may be that an interdental realisation of /T/ is preferred as the subject has a dental realisation of /t/, and a reported tendency to produce /T/ as a stop (Lavoie 2003) would possibly neutralise the /t/ ~ /T/ contrast, were /T/ dental. This possibility remains to be investigated in more detail. Experiment 2 studied the generation of frication in dental and interdental /T/ for the same subject. The role of the lips in the generation was found to be considerable, confirming the comments in Shadle (1997: 47) and contrary to the statements in Laver (1994), Ladefoged and Maddieson (1996), Stevens (1998) and Cruttenden (2001). In all these latter sources, the generation of frication for /T/, whether dental or interdental, was stated to be at the point of constriction. The lower lip appeared to be more important in the generation of turbulence for the dental fricative, and the upper lip was more important in the generation of frication for the interdental fricative. Ladefoged and Maddieson (1996: 138) and Stevens (1998: 249) characterise (inter)dental /T/ as a non-sibilant (or as [-strident], here taken to be equivalent to sibilant pace Stevens 1998, et al.) on the basis of the assumption that /T/ does not involve an