Lab Report Write Up

Instrumental Phonetics: Lab Project

“Why is it called a ‘ham-bag’? It doesn’t carry ham!” Cases of assimilation in Northern Irish speech. Introduction The question posed in the title of this paper is one that was recently asked my one of my (obviously) male friends. The purpose of this study will investigate how this comedic ambiguity arose, discussing how the place of articulation of the word-final alveolar segment is affected by the place of articulation of the following word-initial segment. I wish to investigate to what degree this ‘assimilation’ occurs in my own speech, and whether fast or slow speech rate affects this. To avoid any ambiguity, I will adopt the term ‘assimilation’ to refer to any instances of one segment becoming more like another, encompassing the range of potential factors at play from deletion to reduction. I suggest this at the beginning to avoid any of the uncertainty correlated with the differing usages within the literature. Within the investigation, I also wish to examine whether any evidence can be seen to support varying degrees of assimilation, or to suggest that it is a process of gestural overlap that is affecting the segments under analysis. I will also discuss the notion of phonological representation and, contrastively, phonetic implementation. While the notion of assimilation is widely discussed within the field, ‘articulatory phonology’ presents an opposing analysis of what occurs in the processes of connected speech. Traditional assimilation theory argues for cases where “two distinct underlying segments abut, and one “adopts” characteristics of the other to become more similar, or even identical, to it.” (Nolan p. 262). Conversely, Browman & Goldstein claim that cases of apparent assimilation are really due to gestural overlap; where the “basic units” of phonological contrast are known as “gestures” which are abstract characterizations of natural classes of sounds, encompassing both duration and time. (Browman and Goldstein, 1992. p. 160) The conflicting theories that are predominant within phonology affect the predictions made about the following experiment. If the results suggest complete assimilation of the segments, this may provide more evidence for the former theory of assimilation. However, if the majority of the evidence suggests segments becoming more like each other, rather than actually ‘becoming’ each other, it may enable us to argue more for the theory of gestural overlap. They also argue for different possibilities of what could be going on in the mental and physical domains of speech. While the traditional view (supported by SPE) posits a great deal of influence coming from the abstract mental phonological representations and consequent processes, the gestural approach puts less emphasis on the

abstractness of these representations. It does, however, suggest a mental representation that contains a temporal element. By analysing the effects on the word-final alveolar, I will attempt to relate any findings to the theories presented. To carry out the study, a recording was made of my own voice. I recorded 4 repetitions of 6 test sequences at a careful speech rate and then 4 repetitions of the same sequences at a fast speech rate. The recordings were saved as sound files, and by using acoustic analysis software called Praat (Boersma & Weenink, 2009), the speech was analysed to uncover what is happening to the segments under discussion. Barry (1985) and Kerswill (1985) were also both interested in the effect of such connected speech processes. While they used EPG recordings for their analysis, the findings provide some input on what could be expected from the present study. They found that in general, there was a tendency to “make less alveolar contact in the faster tokens”. While this means that the alveolar was not fully realised in these cases, it does not mean that there was complete assimilation to the following velar. The prediction, however, that can be made from this is that cases of assimilation should be more evident in fast speech than slow speech. (Nolan p. 264) Another prediction is that the results may vary among the different lexical sets as different vowels are used. “…differences in phonological form will always result in distinct articulatory gestures.” At this stage I wish to highlight that the study makes within-speaker comparisons and that the regional dialect may also affect certain factors. (Nolan, p. 272) I predict that cases of assimilation will be evident as a result of previous research, but also because of the question posed in the title. It is a phenomenon that we seem to be aware of. The title of the paper, however, raises some issues for how this incident occurs. Nolan questions whether residual alveolars are sufficient to cue the perception of a lexical alveolar. In the title, this does not appear to have happened. This addresses the notion of a gradual articulation process as in this case, an articulatory continuum of forms has not been productive for conveying the meaning to the listener. Method The study was carried out on my own voice. I am a 20 year old female, speaking with Belfast Vernacular English. A common characteristic of Northern Irish (female) speech is that it tends to be quite fast, and I am often told this about my own speech. Therefore, it will be interesting to see the results given by the different speech rates. The experimental materials consisted of 3 pre-designed question-answer sequences; 1. What kind of gadget was it? It was a fad gadget. 2. What kind of gadget was it? It was a fag gadget. 3. What kind of tablet was it? It was a fad tablet.

I then designed my own test materials that had to fit the following criteria; • • • •

A word-final /d/ following a vowel Preceding word-initial alveolar and velar stops A word-final /g/ following the same vowel preceding a word-initial velar stop There must be a back vowel at the beginning of the test sequence

The idea was for the test sentence to be in as comparable a context as possible. There needed to be the same number of syllables in the test sequences, with phrasal stress falling on the default position because of the question format of the background sentence. After ensuring the above criteria were met, the following materials were decided upon; 1. What kind of table was it? It was the dud table. 2. What kind of cable was it? It was the dug cable. 3. What kind of gable was it? It was the dug gable. When the experimental materials were ready, it was time to make the recordings. In the studio we followed three steps that are used by all sound capturing equipment to make high quality audio recordings of the human voice for analysis. Capturing: Within the studio, there were two rooms. An isolation booth where an AKG CK 98 hypercardoid microphone was used to record the test sentences. This kind of microphone is appropriate for capturing human voice as it is highly directional and rejects sound from everywhere, except from directly in front of it. There was also control room were the technical elements of the recording are controlled. Encoding: At this stage, a recorder inscribes the electrical signal into a device called MOTU 828. This device is used to control both the recording level and the volume of the recordings. It is also an analogue to digital converter (ADC) that encodes the electrical signal as a binary code which is stored on the computer as an audio (wav) file. A piece of software called SONAR makes sense of the wav files and allows you record, playback and edit them. Playback: Where the digital information is converted back into electrical signal. The sound is then played back through the speakers. The sampling frequency that we used was 48 kHz, and the bit depth was of 16 bit. This ensures a high quality recording as a higher sample rate allows a more accurate representation of the original sound. Once the sound files had been saved onto a computer system, they were analysed using acoustic analysis software called Praat. (Boersma & Weenink ,2009). Three F2 measurements were taken from the final three pitch peaks of the vowel preceding the assimilation site. The test sequences were repeated four times at each speech rate to allow generalisations to be made as both F1 and F2 values can differ for the same vowel.

When the F2 readings had been recorded, the means and standard deviations were calculated and these will now be presented within the results. Results The results can be observed in the following graphs.

Fad Gadget vs. Fag Gadget 2500

Frequency (Hz)

2000 1500

Fag Gadget Fad Gadget

1000 500 0 Slow

Fast Speech rate

Graph 1: Assimilation of post-lexical alveolars and velars ‘fad/gadget’ and ‘fag/gadget’ respectively. Graph 1 shows that the /d/ and /g/ remain somewhat distinct in the slow speech rate, suggesting that in this condition they do not assimilate. The frequency of the alveolar in this condition is similar the frequency readings of the alveolars in the control condition (see Graph 3) for slow speech. In the fast speech condition, the alveolar gets higher in frequency and does not match the control conditions. This graph shows how the frequency reading for the fast speech of the alveolar is almost exactly the same as the fast speech frequency of the target velar. This suggests that complete assimilation has occurred in fast speech. To uncover whether any assimilation can be observed in slow speech, we need to look more closely at the results. The following graph will show the F2 readings for the final three peaks of the preceding vowel for each speech rate.

F2 values for mean slow speech 2500

Frequency (Hz)

2000 fag gadget

1500

fad gadget 1000

fad tablet

500 0 1

2

3

F2 Values

Graph 1a: Mean F2 readings for slow speech rate for ‘fag gadget’, ‘fad gadget’ and control setting ‘fad tablet’. The graph above suggests that in the slow speech rate, the alveolar is partially assimilating to the velar. The F2 readings for the alveolar are getting higher in frequency towards the end, thus becoming more like the velar. This provides some interesting insight to the assimilation/gestural overlap debate and I will talk about this in greater detail within the discussion. The following graph shows the average F2 readings in Dug Cable and Dud Gable.

Dug Cable vs. Dud Gable 1950

Frequency (Hz)

1900 1850 1800

Dug Cable

1750

Dud Gable

1700 1650 1600 Slow

Fast Speech rate

Graph 2: Assimilation of post lexical alveolars and velars in ‘dud/gable’ and ‘dug/cable’ respectively. In Graph 2, we do not see any assimilation patterns. The data used to plot this graph encompasses the averages of all three vowel-final F2 readings and so it may be more appropriate to look more closely at this data.

By looking more closely at the F2 readings of the preceding vowels, we may be able to understand better what is going on in this case. The following graphs show the F2 readings for both slow and fast speech for each test sequence.

Dug Cable 1900

Frequency (Hz)

1850 1800

Mean slow speech Mean fast speech

1750 1700 1650 1

2

3

F2 Readings

Graph 2a: Shows the final three F2 readings for the preceding vowel in the test sequence ‘Dug Cable’. For ‘Dug Cable’, we can see how in both fast and slow speech, the frequency gets lower as the vowel approaches the velar. However, the fast speech condition drops more substantially in frequency than that of the slow speech.

Dud Gable

Frequency (Hz)

2000 1950 1900 1850 Mean slow speech

1800 1750

Mean fast speech

1700 1650 1600 1550 1

2

3

F2 Readings

Graph 2b: Shows the final three F2 readings for the preceding vowel in the test sequence ‘Dud Gable’. This graph shows the sequence in which we would have predicted to see assimilation. In the slow speech condition, the F2 stays at a reasonably constant frequency. In the

fast speech condition, the F2 starts to go up, but then starts to drop as though it is following a similar pattern to those shown in Graph 2a. This may suggest a case of partial assimilation and shows the importance of closely examining the data. In order to validate the results, standard deviations were calculated. Table 1: Standard Deviations of final three F2 measurements in fast and slow speech rates for ‘Dud Cable’ and Dug Gable’. Dud Gable

Slow

78.9

48.2

15.1

295.7

275.1

221.3

85.5

151.9

134.5

224.3

173.1

153.4

Fast Dug Cable

Slow Fast

In the table above, we can see how the standard deviations vary among the speech rates. In ‘Dud Gable’, the standard deviation is relatively small; suggesting that each result is similar to the mean and therefore this is a reliable result. For the faster speech, however, the standard deviations are much larger, demonstrating that the results are much more sporadic and so these results may not be reliable. This would account for what is demonstrated on Graph 2.

Dud Table vs. Fad Tablet 2500

Frequency (Hz)

2000 1500

Dud Table Fad Tablet

1000 500 0 Slow

Fast Speech Rate

Graph 3: Graph to show the control setting of post-lexical assimilation of alveolars. Graph 3 is a representation of the controls that were used. They show the averages of the alveolars in fast and show speech in two different settings. The differences that

can be observed among the two sequences can be attributed to the fact that the vowels are different in terms of backness. The vowel in ‘fad tablet’ (and the other sequences containing this vowel) are realised with the front, low vowel /a/, whereas in ‘dud table’ (and its relative sequences), the vowel is the open-mid, back vowel /ʌ/. The front vowel /a/ gives a lower set of F2 values in the control setting because front vowels are lower with following alveolar rather than velar segments. Back vowels, however, have higher F2 readings preceding an alveolar because of the more drastic movement of the tongue from the back of the mouth to the alveolar ridge. The use of different vowels may have been a factor in yielding different results. In some of the cases, it was difficult to determine the exact location of the final three pitch peaks. This may have resulted in some anomalies within results. To rectify this for future studies, it may be better to either keep similar vowels in each test sequence, or to take a larger sample size in order to provide a more accurate generalisation. Discussion Overall, the results provided some enlightening data. In the first test-sequence, we can see how complete assimilation has occurred within fast speech, and partially in the slow speech condition. In the second test-sequence, while at first there did not appear to be any occurrences of assimilation, on a closer analysis, we saw a hint of partial assimilation of the alveolar in the fast speech condition. This, however, is proposed hesitantly as the standard deviations suggested that the results may not be within reliable confidence limits. In the introduction, I highlighted the purpose; to uncover to what degree word-final alveolars assimilate to following velars. Using specialised acoustic software and spectrogram readings, I was able to take a much closer and reliable look at a real-life speech recording and thoroughly analyse what processes are occurring underneath the level at which we perceive the speech sound. Using fast and slow speech rates gave a comparable setting to uncover the effect of potential connected speech processes. It also enabled a way to relate the findings to well established literature in comparing the settings. The results demonstrate a number of aspects that can be discussed with regards to the hypotheses proposed in the introduction. In the first set of results (‘fad gadget’ vs. ‘fag gadget’) we can observe a case of apparent complete assimilation in the fast speech setting. Using alternative techniques would enable more clarification on the matter but using this particular method suggests the alveolar has been deleted before the velar. Employing the use of Electromagnetic Articulatography (EMA) would enable future researchers to uncover whether or not the tongue-tip is moving at all, therefore proving complete assimilation if it does not. For now, I will argue for complete assimilation based on my own results.

This, therefore, provides evidence for the theory of phonological implementation as one segment becomes another in a specific environment. Consequently it would appear that these phonological processes have derived specific surface representations that have served as an input for the motor commands controlling what my articulators produced. This evidence is favoured by Ladd and Scobbie whose results suggest that gestural overlap is not, on the whole, a suitable model for assimilation. (Ladd and Scobbie) The second sets of results (‘dud gable’ vs. ‘dug cable’) seem somewhat more unreliable. From what can be interpreted, it appears as though some partial assimilation can be observed. This could also be interpreted as a ‘reduced alveolar’ in that the tongue tip may still be raising, but not enough to make a full alveolar. This finding is supported in Ellis and Hardcastle (2002) where the notion of reduced alveolars is proposed to account for what is being shown in their EPG data. Reduction can also be thought of as reducing the magnitude of the gesture. In this sense, it may be that the magnitude of the gesture is decreased as it overlaps with another gesture. A theory such as this would be popular for researchers such as Nolan, who suggest this to be a much more intuitive way of organising the articulators, and Browman and Goldstien who have also provided evidence for gestural overlap. In the second set of results (Graph 2), I was much more aware of pressures to be clear in pronouncing the separate segments because of the word-initial voiceless velar. As a result of this, the motor commands where to articulate each process more distinctly. In the spectrogram this is very obvious in certain cases where it showed a lot of noise in the voiceless velar, much more than the voiced. This suggests that the articulators where working to intensify this segment. It is also prudent to note that in the second set, the voiced alveolar precedes a voiceless alveolar rather than a voiced one as in dataset one. This extra feature may have skewed the results, especially in the situation in which the recording took place which I will now discuss. I found recording my own voice quite a daunting experience! While there were only 3 fellow students and a technician in the control room, the setting was quite intimidating and hearing my own voice was quite strange! I found it quite difficult to speak in a natural way and found myself tripping over words and my mouth getting very dry. I feel this may have affected the results as it may have changed the true frequencies at which I speak. While trying to concentrate on reading out the sequences, I did not intonate my question as a question should be, and so the sentences sound slightly unnatural. As I am often told that I speak to quickly, in formal situations, I find myself consciously trying to be more careful with my speech rate. Other socio-linguistic factors also tend to be at play on my style of speech. The Belfast accent can be quite broad and so I do tend to annunciate more. This is done in one sense to be understood better, but it is also an attempt to avoid the broader varieties of my accent. I believe that these factors may have majorly influenced the data and it may be more natural to try and analyse speech produced in a more naturalistic setting. While the purpose of making the test sequences was not revealed until after the recording was made, as a linguistics student I found myself speculating over what the

motivation could be. While I attempted to not let this influence my recording, I believe these speculations still affected the way in which I said the sequences within the recording. As a result of this, it may be useful for future studies to make within-subject comparisons over a range of recordings. I know for myself, the results may have been more reliable did I have time to go back and do the recordings a few more times in the studio. In future studies, I believe it would be extremely beneficial to make use of other methods of speech analysis alongside the one used here. While the study allowed generalisations to be made, as I have previously mentioned, it could not be confidently clarified whether the cases of ‘complete’ assimilation where actually ‘complete’ and using other software, a thorough analysis could be made on what the articulators are truly doing.

References Browman, C. P. and Goldstein, L. (1992) Articulatory Phonology, An Overview. S. Karger AG, Basel. Pp. 155-180. Docherty, G. and Ladd, R.D. Laboratory Phonology 2. Cambridge University Press. pp. 261-280

Ellis, L. and Hardcastle, W. J. (2001) Categorical and gradient properties of assimilation in alveolar to velar sequences: evidence from EPG and EMA data. Journal of Phonetics. Ladd, R. D. and Scobbie, J.M. External sandhi as gestural overlap? Counter-

evidence from Sardinian. Boersma, P., & Weenink, D. (2009). Praat: doing phonetics by computer (Version 5.1.04) [Computer Program].