Inter- and Intra-Subject Variability: A Palatometric Study

Transcription

1 Brigham Young University BYU ScholarsArchive All Theses and Dissertations Inter- and Intra-Subject Variability: A Palatometric Study Marybeth Corey Sanders Brigham Young University - Provo Follow this and additional works at: Part of the Communication Sciences and Disorders Commons BYU ScholarsArchive Citation Sanders, Marybeth Corey, "Inter- and Intra-Subject Variability: A Palatometric Study" (2007). All Theses and Dissertations This Thesis is brought to you for free and open access by BYU ScholarsArchive. It has been accepted for inclusion in All Theses and Dissertations by an authorized administrator of BYU ScholarsArchive. For more information, please contact scholarsarchive@byu.edu, ellen_amatangelo@byu.edu.

2 INTER- AND INTRA-SPEAKER VARIABILITY: A PALATOMETRIC STUDY by Marybeth C. Sanders A thesis submitted to the faculty of Brigham Young University in partial fulfillment of the requirements for the degree of Master of Science Department of Communication Disorders Brigham Young University December 2007

3 BRIGHAM YOUNG UNIVERSITY GRADUATE COMMITTEE APPROVAL of a thesis submitted by Marybeth C. Sanders This thesis has been read by each member of the following graduate committee and by majority vote has been found to be satisfactory. Date Christopher Dromey, Chair Date Shawn Nissen Date Ron Channell Date Samuel Fletcher

4 BRIGHAM YOUNG UNIVERSITY As chair of the candidate s graduate committee, I have read the thesis of Marybeth Corey Sanders in its final form and have found that (1) its format, citations, and bibliographical style are consistent and acceptable and fulfill university and department style requirements; (2) its illustrative materials including figures, tables, and charts are in place; and (3) the final manuscript is satisfactory to the graduate committee and is ready for submission to the university library. Date Christopher Dromey Chair, Graduate Committee Accepted for the Department Date Ron W. Channell Graduate Coordinator Accepted for the College Date K. Richard Young Dean, David O. McKay School of Education

5 ABSTRACT INTER- AND INTRA-SPEAKER VARIABILITY: A PALATOMETRIC STUDY Marybeth Corey Sanders Department of Communication Disorders Master of Science Abstract Electropalatometry (EPM) has proven to be a useful clinical and research tool for measuring tongue-to-palate contact. The goal of this study was to determine whether the development of a database of standardized palatometric articulation files is feasible by examining the variability which exists within and between speakers. Twenty standard American English dialect speakers were fitted with palatometer pseudopalates. Test stimuli were VCV nonsense words using a schwa in the initial position, the 15 palatal consonants, and three corner vowels, /ɑ/, /i/, /u/. From these palatometric recordings a variability index was created to examine intra- and interspeaker variability. Different aspects of articulation (i.e., place, manner, voicing, coarticulation) were considered. Significant findings for variability were found for place of articulation in the /i/ vowel context and for manner of articulation in the /ɑ/ vowel context. Also in the /ɑ/ vowel context, significant findings were found between the commonly misarticulated /l/,

6 /r/, and /s/. Consonants coarticulated with /ɑ/ were found to be significantly less variable than consonants coarticulated with /u/. Also, speakers who were more variable in one vowel context, tended to be more variable in other vowel contexts. These quantitative findings, as well as qualitative observations, are discussed from theoretical and clinical perspectives. Directions for future research are outlined.

7 ACKNOWLEDGMENTS I have many to whom I am indebted to for helping me through this project and my master s program. I want to express my gratitude to Dr. Dromey for always making time to hear my constant flow of questions, concerns, and complaints. Also, I want to thank Dr. Fletcher for his expertise in this area and funding of the project, as well as, Dr. Nissen and Dr. Channell, for their advice and encouragement throughout this project. I would like to acknowledge Lynn Tyler, Clinic Director of Dental Assisting at Ameritech College, for making all of the dental impressions. I am also grateful for my willing subjects which included family, friends, and strangers who took many hours out of their own busy schedules to be a part of this study. I am also very grateful to my family and their support: My husband Mike, for encouraging me when I didn t think I could make it through and always letting me know this was as important to him as it was to me; My parents, for always letting me know they were proud; My sister Jill, for telling me I could do anything for 2 years and fighting off five children to edit my paper; My family and in-laws, for being interested in this project even if they didn t always understand it. Last, but far from least, I am grateful towards my Heavenly Father and the loving support I felt from Him through this project and my master s program.

8 vii TABLE OF CONTENTS Page List of Figures... ix List of Appendixes...x Introduction...1 EPM Uses and Findings... 5 Considerations for Research... 7 Purpose of Study... 8 Method...11 Participants Procedures and Stimuli Data Analysis Results...15 Place of Articulation Manner of Production Voicing Gender /l/, /r/, /s/ Coarticulation Description of Consonant Productions Discussion...32 Place of Articulation Manner of Production... 33

9 viii l/, /r/, /s/ Coarticulation Description of Consonant Productions Conclusion References...40

10 ix LIST OF FIGURES Figure Page 1. Example of bimodal distribution of contact patterns across 10 trials from Speaker 1 for /sɑ/ Effects of place of articulation on variability in /i/ vowel context Effect of manner of production on variability in /ɑ/ vowel context Comparison of /sɑ/, /lɑ/, and /rɑ/ variability Within vowel comparison of consonant variability in /ɑ/ vowel context Within vowel comparison of consonant variability in /i/ vowel context Within vowel comparison of consonant variability in /u/ vowel context Scatterplot of correlation of mean VI of speakers across all consonants Demonstration of difference in variability among different speakers Contact across all speakers in /ɑ/ vowel context Contact across all speakers in /i/ vowel context Contact across all speakers in /u/ vowel context Demonstration of increased tongue-to-palate contact for high /i/ and /u/ vowels compared with low /ɑ/ vowel Demonstration of the incomplete closure of velar stops...31

11 x LIST OF APPENDIXES Appendix Page A. Speaker Consent to Participate in a Speech Research Study...45 B. Word List Used During Data Collection...47 C. Variability Index for Consonants in /ɑ/ Vowel Context...48 D. Variability Index for Consonants in /i/ Vowel Context...50 E. Variability Index for Consonants in /u/ Vowel Context...52 F. Average Number of Contacts Across Subjects (out of 10 trials) in /ɑ/ Context...54 G. Average Number of Contacts Across Subjects (out of 10 trials) in /i/ Context...63 H. Average Number of Contacts Across Subjects (out of 10 trials) in /u/ Context...72 I. Gender Effects for /ɑ/...81 J. Gender Effects for /i/...82 K. Gender Effects for /u/...83

12 1 Introduction Dynamic palatometry or electropalatometry (EPM), first described in an article by Fletcher, McCutcheon, and Wolf in 1975, is a technique that uses a custom fit pseudopalate with 60 to 118 embedded electrodes or sensors which allows quantitative measures of tongue-palate contact. When the tongue makes contact with the sensors, a circuit is completed. The pseudopalate sensors contacted during speech are displayed on a computer screen (Bernhardt, Loyst, Pichora-Fuller, & Williams, 2000; Fletcher et al., 1975). A display of tongue-palate contact using this type of technology is called electropalatography (EPG). EPG is a form of palatography, which is a more general term that refers to the study of visual patterns of tongue-palate contact, without the extraction of numeric data. Using tin foil to show the tongue s contact patterns or coating the palate with material that can be removed by contact with the tongue are examples of nonelectronic palatography (Fletcher, 1992). Most auditory training programs for hearing impaired individuals and phonological therapy programs are based on the assumption that perception precedes production (Crawford, 1995). In other words, deaf children need to perceive, in some way, the sound they are being taught before they can produce the sound. Obviously, auditory means of perception are not usually effective in teaching speech production with such disorders as hearing impairment. The benefit of the EPM technology is that it uses visual perception to teach production. In therapy with EPM, clients watch the clinician s model on the computer screen and then try to imitate that model on the screen while receiving immediate visual feedback (Bernhardt et al., 2000; Fletcher et al., 1991). This dynamic, real-time display of articulation is a unique and important feature of speech. The client is able to see articulation that would otherwise be invisible during

13 2 speech, particularly the lingual patterns associated with consonants and non-low vowels (Bernhardt et al., 2000; Dagenais, 1995). EPM can also provides clinicians with a more objective and accurate source of information than an auditory-based transcription, since deviant speech gestures do not always have an audible consequence. EPM can also facilitate contrastive learning by showing the client their incorrect (pre-therapy) productions in comparison to their more correct productions based on palatometric feedback (Bernhardt et al., 2000; Dagenais & Critz-Crosby, 1991). Finally, EPM allows the clinician to measure the effectiveness of therapy both qualitatively and quantitatively (Hardcastle, Gibbon, & Jones, 1991). Qualitatively, a clinician is able to judge whether the therapy had an effect on the client s speech by seeing changes in the speech patterns on the screen. Quantitatively, a clinician can collect data from the visual diagram of the palate. By using the amount or frequency of contact, determined by the sensors displayed on the diagram, a clinician can measure how much change has actually taken place (Byrd & Flemming, 1995). Data reduction methods can contribute to the collection of quantitative data from EPM. Researchers using data reduction may consider a variety of displays using raw data in the form of individual frames of the pseudopalate display. For example, a Totals display divides the palate into regions or zones (i.e., frontal, medial, and back) and the number of electrodes in that region contacted is plotted as a function of time. This compares the proportion of the total number of electrodes that were contacted in each region (Byrd & Flemming, 1995; Hardcastle, Gibbons, & Jones,1991; Hardcastle, Gibbon, & Nicolaidis, 1991). This method can differentiate between phonemes (e.g., /s/ which is more anterior versus /ʃ/ which is more posterior) depending on the amount of

14 3 contact in an area rather than collecting data for specific electrodes. Another example is Frequency of Contact or temporal displays. These present the frequency of contact with each electrode over time, allowing the clinician to observe where on the palate the majority of contact was made (Byrd & Flemming, 1995; Hardcastle, Gibbons, & Jones, 1991). In a 1998 study, Friel used palatographic data to identify underlying abnormalities in tongue contact patterns. The participant s speech problems could not have been identified through auditory means alone. Abnormal articulation gestures could have been reinforced with traditional therapy because they may have sounded normal in monosyllabic words. Friel s study is a good example of how a palatographic system provides important contact information from the display that can help create, along with the clinician s feedback, a methodical, clear therapy approach to obtain clues towards a diagnosis as well as increase intelligibility (Fletcher et al., 1991; Fletcher & Hasegawa, 1983, Hardcastle et al., 1987). EPM can be successful for clients who have had little success with other techniques because it allows a unique approach to treating articulatory problems. EPM allows clinicians to break the therapy process down into components and separate the steps of acquisition (i.e., establishing the place of articulation) then later add other components such as voicing. Clinicians can also plan treatment that directly meets the articulatory needs of the client because specific articulatory data are collected. EPM helps clients develop a better idea of what they need to change to have a more normal production. Lastly, EPM can be inherently motivating to a client because it details the problem with clear, visual feedback. Through this, they can see what they need to do to

15 4 correct their productions. Therefore, this technique actively involves the client in the therapy process (Hardcastle, Gibbon, & Jones, 1991). On the other hand, researchers have expressed some concerns about the limitations of EPM in therapy and research. Firstly, pseudopalates may interfere with sensory feedback and articulation, thus causing abnormal articulatory movements (Byrd & Flemming, 1995). Researchers have found that the pseudopalate does not substantially affect articulatory feedback. Lingual feedback has been shown to compensate for the loss of tactile feedback from the surface of the hard palate (Fletcher, 1992). Specifically, Fletcher (1992) found the pseudopalate has little effect on articulation when it is less than.5 mm thick and closely fits the contours of the teeth and palate. An informal experiment conducted by Hardcastle (1972) found no significant differences between the speech of participants with or without a pseudopalate for speech rate, intensity, and fundamental frequency. Another limitation reported in the literature is sensor placement. Sensors are limited to the hard palate, making recording of soft palate contact, such as back vowels, more vague, although this is not as much of an issue for less posterior velar sounds such as /k/ and /g/ (Byrd & Flemming, 1995; Recasens, 1991). Also, sensors are at discrete points along the palate. The data collected from the sensors does not provide information on whether the contact is continuous for the blank areas between the sensors (Hardcastle, Gibbon, & Jones, 1991). In considering this concern, researchers have found that sensors can be placed as close as 1mm apart without serious problems of saliva bridging between them and shorting out the contact. An array of 96 sensors distributed 3- to 5-mm apart

16 5 has been found to be adequate for most articulation measures (Fletcher, 1992; Hardcastle, 1972). A further limitation is that EPM only records contact and timing information. Trajectory, tongue shape, articulatory velocity, proximity of tongue to palate, the part of the tongue contacting the palate, and articulations that require no tongue-palate contact are not recorded (Byrd & Flemming, 1995; Goozee, Murdoch, & Theodoros, 2003; Hardcastle, 1972; Hardcastle, Gibbon, & Jones, 1991; Hardcastle, Gibbon, & Nicolaidis, 1991). Although the Logometrix EPM system does provide some labial information, in general, EPM is not useful in the remediation of articulation disorders involving lingualdentals, glottals, and some vowels. Lastly, individual variations in palatal anatomy make it difficult to design universal pseudopalates. Each participant must be fitted with an individualized pseudopalate with a somewhat individualized sensor placement. Due to the uniqueness of an individual palate and the cost of making pseudopalates, it is difficult to conduct comparative experiments with large sample sizes which would accurately represent a population (Bryd & Flemming, 1995; Hardcastle, 1972; Hardcastle, Gibbon, & Jones, 1991). EPM Uses and Findings The palatometer has been used in remediation and research with a number of communication disorders. Several studies have demonstrated the palatometer s utility in treating speech deficits associated with hearing impairment (Bernhardt, Gick, Bacsfalvi, & Ashdown, 2003; Bernhardt et al., 2000; Crawford, 1995; Dagenais, 1992; Dagenais & Critz-Crosby, 1991; Fletcher et al., 1991; Fletcher & Hasegawa, 1983). Others have examined its application with esophageal speakers (Christensen, Fletcher, &

17 6 McCutcheon, 1992), developmental articulation/phonological disorders (Dagenais, 1995; Friel, 1998; Hardcastle, Gibbon, & Jones, 1991; Hardcastle et al., 1987), acquired speech disorders (Goozee et al., 2003; Hardcastle, Gibbon, & Jones, 1991), and cleft palate speech (Hardcastle, Gibbon, & Jones, 1991). Additionally, the palatometer has been used for accent reduction (Bright, 1999). Hardcastle (1972) also suggested foreign language learning and oral physiology research as other uses for EPG. Bernhardt et al. (2003) found that when using an EPM unit, clients tended to make more progress with phonemic targets that were absent or marginal in their speech before therapy than with phonemic targets that were more established before therapy. Similar results were also found by Fletcher et al. (1991). Furthermore, these results are consistent with Dagenais (1992) claim that visual feedback may be most effective in establishing new articulatory patterns. Profoundly hearing-impaired unintelligible children were able to learn significantly better sound production skills in a shorter time than with the traditional therapy they had undergone before Dagenais study. Dagenais found that although both traditional training and physiological (EPG and glossometry) training were beneficial, children using the physiological techniques tended to learn more accurate sound production skills and also had increased retention of the sounds they learned. Fletcher (1989) examined stop, affricate, and sibilant production in CV syllables spoken by normally speaking children using an EPG system. He analyzed the movement towards constriction, consonant sound production, and vowel sound production. Fletcher noted that the tongue contact for affricates and stops remained fairly consistent for a time after the rise to its peak articulatory position. The reason for this was likely due to the

18 7 need to build up pressure for sound production. This is also why more sensors were activated with stops and affricates than with sibilants. He also noted that a momentary plateau in the tongue and palate release in affricates likely represents the stop-to-sibilant transition in the production of an affricate. Another interesting finding was that the voiced members of the sibilant groups (i.e., /s/ and /z/) have.75 to 1 mm narrower grooves than their unvoiced counterparts. Additionally, Fletcher found significant age effects in that the older speakers had a shorter response time, their place of contact was more posterior than the younger participants, and the area of contact was reduced. Considerations for Research In general, most people make the same sounds in approximately the same way, although the exact articulator placement may vary from one person to the next (Fletcher, 1992). For this reason, normal speech is also considered variable, especially in its timing structure. This creates problems in the instrumental assessment of pathological speech. Variations in the way a phoneme is articulated will become apparent when varying the neighboring phonemes, the tempo, or the stress pattern. Even the degree of variability will differ from one phoneme to another (Butcher, 1989). Specifically, McAuliffe, Ward, and Murdoch (2001) looked at variability of consonant production in the /i/ vowel context and found that /l/ had a significantly higher degree of intra-speaker variability than did /t/ and /s/. In fact, /t/ was significantly more stable than /l/ within speakers. The velar stop /k/ was also found to be more stable than /l/, but not significantly so. Cheng, Murdoch, Goozee, and Scott (2007) looked at the effect of age on variability and found that the 6- and 7-year old children in their study were significantly more variable in their tongue contact patterns for the velar stop /k/ than the 12- to 17-year olds and the adults. With regard to disorders, Dagenais and Critz-Crosby (1991) found

19 8 that normally hearing participants tongue-palate contact patterns for consonants were more consistent than those of the hearing impaired participants. Another issue to consider in research is the choice of vowels. Most researchers choose from corner vowels, often using /i/ in CVC or CV words. An /i/ has been shown to have a low degree of coarticulartory variability in palate contact (Goozee et al., 2003; McAuliffe et al., 2001; Recasens, 1991). Dagenais and Critz-Crosby (1991) found that for normally hearing participants, velar stops were the only consonants that showed a coarticulatory effect with the vowels /i/ and /ɑ/. The contact for the velar stops moved anteriorly when they were coarticulated with an /i/. This is in agreement with the findings of Fletcher (1989). On the other hand, Stone, Faber, Raphael, and Shawker (1992) found that although post-consonant vowels had minimal effect on the consonants, prevocalic consonants had a substantial effect on following vowels. However, /i/ was shown to have a significantly greater effect than /ɛ/, /ɑ/, and /o/ on prevocalic consonants. Stone et al. also noted that /i/ had the greatest palatal contact compared to other vowels. Later research by Dagenais et al. (1994) found that alveolar stops were influenced both by the vowel and by voicing, although the contact patterns were affected differently depending on the context. Purpose of Study Butcher (1989) stated that quantitative descriptions of normal dynamic tongue contact patterns should eventually be possible, firstly, to establish what constitutes the normal range of speech behavior and, secondly, to measure the degree of deviance from this norm (p. 47). This is consistent with other researchers statements concerning

20 9 the need for data on normal articulation patterns to better judge abnormality (see also, Higgins, Netsell, and Schulte, 1994; McAuliffe et al., 2001). Currently the Logometrix palatometer software allows speech pathologists to save files containing personal palatometric data so that clients can practice at home. The clinician could also use such a file to quantify their client s progress toward more normal patterns and identify normal and abnormal articulatory postures. Over a course of therapy, a clinician might adapt to the client s speech and may not be able to accurately judge whether there was an actual increase in intelligibility or just an increase in the clinician s ability to comprehend the disordered speech. Given standardized articulatory palatometric data for comparison with clients personal palatometric files, this perceptual bias can be overcome. Fletcher (1989) stated that the search for commonalities in speech production logically begins with maneuvers that normal talkers use to produce standard sounds of language. A range of functionally equivalent articulatory actions may then be identified and their common characteristics defined (p.736). A normative database of standardized palatometric articulation files could meet many clinical and research-related needs. Besides the problem of the clinician s perceptual bias in tracking client progress in the absence of instrumentation, the lack of standardized palatometric data creates other issues for both clinical and research applications. In the clinic, anatomic differences in the therapist s mouth could leave room for error when using his or her own speech as an EPM model in therapy. The therapist could be providing the client with a model that is atypical. As far as research is concerned, there is value in a standardized database that allows comparisons with subsequent data sets. Studies in the past have attempted to create their own non-

21 10 disordered data set for comparison to disordered groups. But without a standard norm across studies, researchers cannot readily draw inferences and compare their findings with previous research. A standardized database would help to alleviate these issues. The goal of this study is to determine whether the development of a database of standardize palatometric articulation files is feasible. Could a database be developed representing optimal articulation patterns? This study will examine standard American English dialect speakers and their inter- and intra-individual variability to consider whether the variability is small enough to develop normative palatometric files.

22 11 Method Participants Twenty participants between the ages of 18 and 35 took part in the study. The 10 male (M = 25.9 years) and 10 female (M = 24.4 years) participants were recruited by announcements and fliers. They reported no history of speech abnormalities, jaw problems, hearing impairment, or serious dental abnormalities. Each participant passed a hearing screening bilaterally at 500, 1000, 2000, and 4000 Hz at 15 db HL. Procedures and Stimuli Each of the participants had dental impressions made, from which plastic palatometer pseudopalates were created. Before the recording, the participant was put through a desensitization process where they engaged in conversation for at least 30 minutes in order to adapt to the presence of the pseudopalate in their mouth. The recordings were made in an Acoustic Industries 7 x 7 single walled sound booth using a head-mounted condenser microphone (AKG C-420) for the palatometric acoustic recording placed approximately 4 cm from the mouth for each recording. The LogoMetrix EPM system used in this study, which has 118 sensors, tracked the tongueto-palate contact. Data were collected for controlled, isolated word productions. The participants were asked to repeat a variety of VCV nonsense words ten times. The initial vowel for each word was a schwa in order to place the tongue in a neutral position before each consonant production. Fifteen consonants were used (/t, d, k, g, s, z, n, ʤ, ʧ, ʃ, ʒ, l, r, j, ŋ/), each in the context of 3 corner vowels, /i, ɑ, u/ in the final position of the VCV nonsense word. An exception was made for /ŋ/ since it is not used in initial position in the

23 12 English language. /ŋ/ was recorded in a VC context using the corner vowels. The nonsense words used during the recordings can be found in Appendix B. The Logometrix Palatometer software recorded which sensors were contacted during each utterance of the VCV nonsense words. Data Analysis The maximal contact frame (the instant at which the highest number of electrodes were activated) was used as the representative frame for each consonant production. In cases where the maximal contact was not representative of the consonant, namely with the high vowels /i/ and /u/, the researcher scanned the frames and chose a representative frame. A number of productions were discarded from the data set due to either recording instrument difficulties or misarticulations of the target sound. Misarticulations included perceptually deviant articulations (i.e., producing a /z/ in the place of /s/), an unnaturally long pause between the schwa and the consonant, and the complete deletion of the proceeding schwa. For each speaker for each consonant, a total was calculated for each electrode. This total represented the number of times across ten repetitions that a given electrode was activated by tongue contact. For those consonant trials with fewer than 10 acceptable tokens, the numbers were averaged and then recalculated to be out of 10 to allow consistency in data analysis. With the totals calculated, a chart was created to record the number of ones, twos, threes, and so forth up to tens found from the list of totals for each electrode. A bimodal distribution was expected, because consistent repetitions would lead to a score of 10 (electrode contacted for every repetition) or 0 (electrode never contacted during any repetition). Any total other than 0 or 10 reflects a degree of inconsistency in

24 13 the articulatory contact for that sound. Figure 1 is an example of the data that were used for this process. Variability was calculated for each consonant for each participant (see Appendix C through E). A variability index (VI) was calculated by taking the ratio of the number of totals that were not zeros or 10s to the total number of sensors (118). The total contact for each consonant was used to find a mean number of contacts for each electrode. In the tables in Appendix F through H, the numbers given are out of 10, giving a ratio of contact across all speakers and all trials across vowels. The traditional anatomical definition of place was considered. In order to examine the effect of place of articulation on variability, the VIs for /t/ and /k/ were compared as were those for /d/ and /g/. Both sets of consonants share a manner of production (stops) as well as voicing characteristics. The consonants differ in the place of articulation. The /t/ and /d/ are alveolar stops and the /k/ and /g/ are velar stops. In order to assess the effect of manner of production on variability, the VIs for /d/, /z/, /l/, and /n/ were compared. All of the consonants were articulated at the alveolar ridge and were all voiced, but they all differed in manner of production: /d/ is a stop, /z/ is a fricative, /l/ is a liquid, and /n/ is a nasal. In order to determine the effects of voicing, the VIs for conjugate pairs were compared against each other (i.e., /t/ and /d/, /k/ and /g/, /s/ and /z/, /ʤ/ and /ʧ/). Gender was considered across the consonants to determine whether it had an effect on the variability of production. VIs for /r/, /l/, and /s/ were also considered due to their clinical significance and frequent misarticulation within the English speaking population.

25 14 80 Frequency Bins Bins Frequency Total 118 Figure 1. Example of bimodal distribution of contact patterns across 10 trials from Speaker 1 for /sɑ/. The number of zeros is the count of electrodes that were never contacted across the 10 trials and the number of tens is the count of electrodes that were contact for every trial.

26 15 Results Place of Articulation A repeated measures ANOVA revealed a significant difference for VI in the /i/ context between /t/ and /k/, F(1,19) = , p =.001, as well as /d/ and /g/, F(1, 19) = p <.001 (See Figure 2). The alveolar sounds were significantly more variable than their velar counterparts in the /i/ context. No significant differences were found for place of articulation in either the /ɑ/ or /u/ vowel contexts. Manner of Production A repeated measures ANOVA demonstrated a significant effect in the /ɑ/ vowel context between /d/, /z/, /n/, and /l/, F(2.294, ) = , p <.001. These results were followed up by a series of paired-samples t tests. These tests showed that there was a significant difference in variability between the sounds: /dɑ/ was more variable than /zɑ/, t(19) = 3.564, p =.002; /nɑ/ was more variable than /dɑ/, t(19) = , p =.028; /nɑ/ was more variable than /lɑ/, t(19) = 3.989, p =.001; and /nɑ/ was more variable than /zɑ/, t(19) = , p <.001. There was not a significant difference in variability between /zɑ/ and /lɑ/, t(19) = , p =.460 or /dɑ/ and /lɑ/ t(19) = 1.713, p =.103 (See Figure 3). No significant effects were found in the /i/ or /u/ vowel contexts. Voicing No significant effects were found when comparing a voiced consonant with its unvoiced conjugate sound.

27 Variability Index ti di ki gi Figure 2. Effects of place of articulation on variability in the /i/ vowel context.

28 Variability Index dɑ zɑ nɑ lɑ Figure 3. Effect of manner of production on variability in the /ɑ/ vowel context.

29 18 Gender A few gender effects were noted from a series of one-way ANOVAs, but since there were a large number of tests run on the data, it increases the risk of a type 1 error. For this reason, it was difficult to determine with confidence which findings were truly significant. See Appendix I through K for the one-way ANOVA test results. /l/, /r/, /s/ A repeated measures ANOVA demonstrated a significant effect in the /ɑ/ vowel context between /r/, /l/, and /s/, F(2, 38) = 3.569, p =.038. These results were followedup by a series of t tests which showed that /sɑ/ was significantly less variable than /rɑ/, t(19) = , p =.005. /lɑ/ was not significantly more or less variable than either /rɑ/ or /sɑ/ (See Figure 4). No significant differences were found in the /i/ or /u/ vowel contexts. Coarticulation An ANOVA using the mean variability across all consonants for each vowel context showed a significant effect, F(2, 38) = 3.285, p =.048. Contrast analyses run concurrently with the ANOVA revealed that productions in the /ɑ/ vowel context were significantly less variable than productions in the /u/ vowel context, F(1, 19) = 8.938, p =.008. Figures 5, 6, and 7 show the variability of each consonant within a vowel context compared with the other studied consonants. The figures show that the patterns of variability were different across the vowel contexts. Although not statistically significant, several qualitative observations are offered. /n/ is consistently one of the most variable sounds across the vowel contexts. The liquid, /l/, tends to be more

30 Variability Index sɑ lɑ rɑ Figure 4. Comparison of /sɑ/, /lɑ/, and /rɑ/ variability.

31 Variability Index tɑ dɑ kɑ gɑ sɑ zɑ nɑ ʤɑ ʧɑ ʃɑ ʒɑ lɑ rɑ jɑ ɑŋ Figure 5. Within vowel comparison of consonant variability in the /ɑ/ vowel context.

32 Variability Index ti di ki gi si zi ni ʤi ʧi ʃi ʒi li ri ji iŋ Figure 6. Within vowel comparison of consonant variability in the /i/ vowel context.

33 Variability Index tu du ku gu su zu nu ʤu ʧu ʃu ʒu lu ru ju uŋ Figure 7. Within vowel comparison of consonant variability in the /u/ vowel context.

34 23 variable in high vowel contexts. The alveolar stops are consistently more variable than the velar stops. The fricatives, /s/, /z/, /ʃ/, and /ʒ/ were the least variable in the low vowel context, /ɑ/, but only the palatal fricatives /ʃ/, and /ʒ/ are the least variable consonants in the /u/ context. The velar stops, /k/ and /g/ were some of the least variable in the high vowel context, but not in the low vowel /ɑ/ context. The scatter plots in Figure 8 show the association between the mean VI across all consonants for each speaker, comparing the different vowel contexts: /ɑ/ and /i/, r =.757, p <.001, /i/ and /u/, r =.862, p <.001, and /u/ and /ɑ/, r =.906, p <.001. This reveals that a speaker who has high variability for consonants in one vowel context is likely to also have high values for the other vowels. This indicates that some speakers are generally more variable than others. An example of this is seen in Figure 9. The electrodes are shaded to show variability and percentage of contact. Darker shaded circles reflect a higher average frequency of contact. A white marker would indicate no contact, and a black marker reflects consistent contact across trials. In this figure Speaker 1 shows a high level of consistency across repetitions, compared with Speaker 11 whose productions were much more variable for the same sound. Description of Consonant Productions Figures 10, 11, and 12 demonstrate the consonant productions in the different vowel contexts as a composite of all of the speakers across all ten trials. Again in this figure, the darker shaded circles represent a higher frequency of contact. Some qualitative observations of contact patterns include the following. The speakers tended to make more tongue to palate contact for most consonants in the high vowel contexts, particularly /i/. This was most notable for posteriors phonemes (e.g., /k/,

35 24 /g/, and /ŋ/). There was more anterior-lateral contact in the high vowel context for velar phonemes. For an example of this increase in tongue-to-palate contact see Figure 13. /r/ also demonstrated increased contact for high vowels, particularly in the /i/ context; /r/ had increased posterior-medial contact as well as anterior-lateral. Incomplete closure was noted for the velar phonemes particularly in the /ɑ/ and /u/ vowel contexts. Some examples of this incomplete closure are seen in Figure 14.

36 25 high /ɑ/ e /i/ e /u/ e low /ɑ/ /i/ /u/ high Figure 8. Scatterplot of the mean VI of all speakers comparing each vowel across all consonants.

37 26 Speaker 1 Speaker 11 /ki/ /ki/ Figure 9. Demonstration of the difference in variability among different speakers. Each palate display represents articulation of /k/ in the /i/ vowel context over 10 repetitions. Darker circles represent more frequent contact across repetitions.

38 27 /tɑ/ /dɑ/ /kɑ/ /sɑ/ /zɑ/ /gɑ/ /ʃɑ/ /ʒɑ/ /ɑŋ/ /nɑ/ /ʧɑ/ /ʤɑ/ /lɑ/ /rɑ/ /jɑ/ Figure 10. Contact across all speakers in /ɑ/ vowel context. Darker shades indicate an increased frequency of contact.

39 28 /ti/ /di/ /ki/ /si/ /zi/ /gi/ /ʃi/ /ʒi/ /iŋ/ /ni/ /ʧi/ /ʤi/ /li/ /ri/ /ji/ Figure 11. Contact across all speakers in /i/ vowel context. Darker shades indicate an increased frequency of contact.

40 29 /tu/ /du/ /ku/ /su/ /zu/ /gu/ /ʃu/ /ʒu/ /uŋ/ /nu/ /ʧu/ /ʤu/ /lu/ /ru/ /ju/ Figure 12. Contact across all speakers in /u/ vowel context. Darker shades indicate an increased frequency of contact.

41 30 /kɑ/ /ki/ /ku/ Figure 13. Demonstration of increased tongue-to-palate contact for high /i/ and /u/ vowels compared with low /ɑ/ vowel.

42 31 /kɑ/ /ku/ /gɑ/ /gu/ Figure 14. Demonstration of the incomplete closure of velar stops.

43 32 Discussion The purpose of this study was to examine inter- and intra-speaker variability in lingua-palatal contact patterns in order to better understand speech patterns of normal speakers. Different aspects of articulation (i.e., place, manner, voicing, coarticulation) were considered when exploring what might have an effect on speaker variability. Significant findings for variability were found for place of articulation in the /i/ vowel context and for manner of articulation in the /ɑ/ vowel context. Also in the /ɑ/ vowel context, significant differences in variability were found between the commonly misarticulated sounds /l/, /r/, and /s/. Consonants coarticulated with /ɑ/ were found to be significantly less variable than consonants coarticulated with /u/. Also, speakers who were more variable in one vowel context tended to be more variable in other vowel contexts. These significant findings, as well as qualitative observations, will be considered and discussed below from theoretical and clinical perspectives. Place of Articulation A possible explanation for /k/ and /g/ being less variable than their alveolar counterparts, /t/ and /d/, in the /i/ context may lie in the fact that /i/ is a high vowel, and coarticulation may account for the differences in variability. The /k/ and /g/ are produced in the posterior part of the mouth and with the posterior lateral tongue contact made during /i/, the tongue may be well positioned to make the velar sounds without first making large articulatory movements. In contrast, for the alveolar sounds the tongue had to move further in the mouth after the /t/ or /d/ to make the appropriate contacts for the following /i/.

44 33 The alveolar stops /t/ and /d/ continued to be more variable than the velars in all of the vowel contexts, although not significantly. This may be due to the part of the tongue being used during sound production. The /k/ and /g/ are produced by using the body of the tongue, whereas, the /t/ and /d/ are produced using the tongue blade to stabilize and the tongue tip for fine articulation maneuvers. Since the body of the tongue is larger in mass, its contact patterns are more consistent, possibly due to less variability in the area it can easily contact. The tongue tip, in contrast, is smaller, more agile and may have a greater potential for variability in the region it can contact during speech. Manner of Production Of the voiced alveolar consonants /dɑ/, /zɑ/, /nɑ/, and /lɑ/, the nasal /nɑ/ was the most variable. The /n/ may be less precise because a complete anterior vocal tract closure is all that is required since the nasal resonance is a primary source of the phonetic differentiation. The exact placement of the tongue to make a /n/ does not affect its intelligibility as much as the other consonants as long as the placement is in the anterior portion of the palate. Also, since nasals, such as /n/ are low in intensity because of the open velopharyngeal port, the oral constriction is not required to impound as much pressure as other types of consonants. The /d/ was the next most variable, possibly for similar reasons as /n/. The exact placement of the tongue when producing /d/ is not as vital to its intelligibility as for other consonants, so long as the tongue tip contacts the anterior portion of the palate. The /z/ and /l/ were the least variable of the alveolar consonants. The /z/ is likely less variable because, like its conjugate pair /s/, it requires more precision to create a tightly controlled lingual groove for high frequency turbulence generation to be perceived

45 34 as a correctly articulated sibilant. A slight deviation anteriorly may make an /s/ sound like a lisp and a deviation posteriorly may make it sound like an /ʃ/. This concept of small lingual movements making major acoustic changes during consonant production was discussed in detail by Stevens (1989). This contrasts with the production of /n/, where large variations in lingual movement result in little acoustic or perceptual change. The /l/ was also less variable. However, the reason for this is unclear. It may be due to the tongue shape and the part of the tongue being used during the articulation of the /l/ in the /ɑ/ vowel context. Because the tongue tip is used exclusively for /l/ in a low vowel context, it may increase the required precision of the tongue-to-palate contact. The only significant finding for the variability of these consonants was in the /ɑ/ vowel context. This may be because /ɑ/ is a low vowel and the tongue was required to travel greater distances between the consonant contact and the low vowel than with the high vowel. When coarticulated with the high vowels, the tongue-to-palate contact increases and that may provide additional stability, which influences the amount of variability a phoneme may have. These coarticulatory anchors are not present with the low vowels and this may contribute to the differences in variability. /l/, /r/, /s/ The lower variability for /sɑ/ than /rɑ/ might be attributed to the perceptual impact of minor placement changes, as suggested above; in other words, a misarticulated /s/ is perceptually more noticeable than other sounds because it either sounds disordered if placed too far anteriorly or like /ʃ/ if placed too far posteriorly. In contrast the /r/ is traditionally considered to be more variable because there are two possible ways to

46 35 produce it, either retroflex or back. With the retroflex /r/, the speaker curls the tongue tip back, whereas with the back /r/ the speaker bunches and pulls the tongue posteriorly. These two different types of /r/ production would create different contact patterns, although both result in an acceptable /r/ production. This variability in /r/ production across the speakers may have contributed to the variability in contact found in the data, although there is not clear evidence of this in the data from the present study. The /r/ productions across speakers, while variable, were made in approximately the same way as seen in Figures 10 through 12. McAuliffe and her colleagues (2001) noted, when using subjects from a different dialect region than those in this current study, that /l/ was the most variable consonant in the /i/ vowel context. However, they did not report data for /r/. The results from this study demonstrated similar findings. The /l/ was one of the more variable consonants in the /i/ context just not significantly so when compared against /r/ and /s/. Coarticulation The reason why consonants coarticulated with the /ɑ/ vowel may be significantly less variable than consonants coarticulated with the /u/ vowel is not immediately clear. Both /ɑ/ and /u/ are back vowels, although /ɑ/ is a low vowel and /u/ is a high vowel. This increase in variability may be due to that difference in production height. The other high vowel, /i/, was also more variable than /ɑ/, just not significantly so. Another potential explanation is that since the consonant production was proceeded by a schwa and the /ɑ/ vowel is closer in proximity of production of the schwa, the variability of the medial consonant may be reduced. The tongue s movement between the vowels and the consonant may be more precise when the tongue is making more similar sounds on both

47 36 sides of the consonant. It could also be speculated that the production of the vowel /u/ may vary more and so affect the coarticulation of consonants. The /u/ can be produced with varying amounts of lip rounding and palatal contact without distorting the sound. This variation may influence the coarticulation effect. The liquid /l/ tends to be among the more variable phonemes in high vowel contexts and one of the least variable in the low vowel context. This may be due to the tongue shape and the part of the tongue contacting the palate. When the /l/ is coarticulated with /ɑ/, the most anterior portion of the tongue tip is being used. However, when /l/ is coarticulated with a high vowel, the anterior as well as more posterior portions of the tongue tip may be used during the production. This difference in tongue contact may contribute to the difference in variability between vowel contexts. The velar stops, /k/ and /g/ were some of the least variable phonemes in the high vowel context, but not with the low /ɑ/ vowel. This increase in variability of velar stops when coarticulated with /ɑ/ was also noted by McAuliffe et al. (2001). When the velar phonemes are coarticulated with high vowels, the tongue-to-palate contact is greater, creating an anchor point for the posterior tongue movement, contributing to greater stability during the sound production. The fricatives, /s/, /z/, /ʃ/, and /ʒ/, were the least variable phonemes in the low vowel context, /ɑ/. The reason for this finding is unclear. It may be that the manner in which fricatives are produced when coarticulated with low vowels such as /ɑ/ decreases the variability of tongue-to-palate contact in some way. The significant correlations of the mean VI of the speakers across the vowel contexts demonstrates that a person who was more variable in one vowel context is likely

48 37 to be more variable in other vowel contexts. In other words, some speakers are generally more variable than other speakers. Figure 9 shows ten repetitions of /ki/ by two different speakers. Speaker 1 is distinctly less variable than Speaker 11. Perceptually, both were normal speakers. Although Speaker 1 was female and Speaker 11 was male, the overall difference in variability was not clearly significant along gender lines as shown in Appendix I through K. The large number of tests conducted made it difficult to clearly interpret any potential gender differences because of the possibility of a type 1 error. Other factors not accounted for in this study may have contributed to this difference in precision between speakers. The physiological data from EPM can reveal interesting differences that do not result in perceptually abnormal speech. It may be that some sounds have a wider range of variability that is not perceptually deviant, but not every speaker utilizes this available range of variability. Description of Consonant Productions Consistent with the findings of previous experiments (Dagenais et al., 1994; Goozee et al., 2003; Hardcastle et al., 1987; McAuliffe et al., 2001), the velar sounds /k/ and /g/ did not always exhibit complete velar closure, particularly in the /ɑ/ and /u/ vowel contexts. Some researchers have stated that this incomplete posterior closure is because the point of contact is on the soft palate. In the /ɑ/ and /u/ vowel contexts, the tongue may be pulled back further in order to coarticulate with the back vowels. If the contact is behind the rows of electrodes on the psuedopalate then the tongue would appear to not make a complete posterior closure (McAuliffe et al., 2001). Also, researchers in the field of Parkinson s disease have noted that spirantization, an incomplete stop closure, is not exclusive to hypokinetic dysarthria. They have found

49 38 that some normal speakers do spirantize stops such as /t/ and /k/ fairly frequently (Weismer, 1984). This may also be an explanation for the incomplete posterior closure of velar stops noted in this study. The speakers tended to make more tongue-to-palate contact for most consonants in the high vowel contexts, particularly with /i/. This was most notable for the posterior phonemes /k/, /g/, and /ŋ/, as well as /r/. A comparison of tongue-to-palate contact for /k/ in each vowel context was shown in Figure 13 to demonstrate this increase in contact. This observation agrees with the findings of Butcher (1989), Dagenais et al. (1994), and Fletcher (1989). More specifically in this current study, /k/, /g/, and /ŋ/demonstrated more anterior-lateral contact in the high vowel context. This might be expected as a function of the contact being pulled forward during the high front production of /i/. The /r/ demonstrated increased posterior-medial contact as well as anterior-lateral contact. This increase in contact is likely due to the coarticulation effect of the high vowels, which have more lateral tongue-palate contact, or anchoring for stabilization, than low vowels. Lateral contact required for the production of the high vowels may blend into the preceding consonant, which also requires lateral contact, for a smoother transition to the vowel. Conclusion In summary, this study examined the variability of phonemes across and within normal speakers in order to give insight into the extent of normal variability. Significant differences in variability were found in connection with the manner of production, place of articulation, and vowel context. These findings increase our understanding of the normal range of variation and may thus help us better understand speakers with abnormal