Effects of the second formant on the perception of velarization consonants in Arabic 9783111357393, 9783111000701

Citation preview

EFFECTS OF THE SECOND FORMANT ON THE PERCEPTION OF VELARIZATION CONSONANTS IN ARABIC

JANUA LINGUARUM STUDIA MEMORIAE N I C O L A I VAN WIJK DEDICATA edenda curat

C. H. VAN SCHOONEVELD INDIANA

SERIES

UNIVERSITY

PRACTICA

XXXIX

1968

MOUTON T H E H A G U E • PARIS

EFFECTS OF THE SECOND FORMANT ON THE PERCEPTION OF

VELARIZATION CONSONANTS IN ARABIC by DEAN H. OBRECHT U N I V E R S I T Y OF R O C H E S T E R

1968

MOUTON THE HAGUE • PARIS

© Copyright 1968 in The Netherlands. Mouton & Co. N.V., Publishers, The Hague. No part of this book may be translated or reproduced in any form, by print, photoprint, microfilm or any other means, without written permission from the publishers.

LIBRARY OF CONGRESS CATALOG CARD NUMBER: 67-31217

Printed in the Netherlands by Mouton & Co., Printers, The Hague.

ACKNOWLEDGEMENTS

Grateful acknowledgement of assistance rendered to the writer is due on a number of fronts and on several different bases, covering various segments of the project. The work was supported financially by the National Science Foundation, under grant number G-19349, indirectly (through its support of the Haskins Laboratories) by the Carnegie Corporation of New York, by Lafayette College and the University of Rochester, under faculty research grants. The generosity of the Haskins Laboratories, New York, in permitting the writer to use its equipment, made the project possible. Dr. Franklin S. Cooper and others of his staff went far out of their way to allow an itinerant researcher to make maximum use of his time during irregular visits to the laboratory over a period of several years. Special thanks are due to Dr. Leigh Lisker, of the University of Pennsylvania and the Haskins Laboratories, for his extraordinarily capable counsel as adviser during the basic work. Principal informants were Mr. Youssif Barakat, of Tripoli, Lebanon, and Dr. Samir Jahshan, of Beirut. Many others gave lesser amounts of time during the early work. The writer is extremely grateful for the cooperation of the American University of Beirut, the Beirut College for Women, the American University in Cairo, the Linguistic Unit, in Cairo, and the Ministry of Education, in Cairo. The United States Embassies in Beirut and Cairo, especially in the persons of Dr. William Cowan and Mr. George Wishon, and Mr. Richard Driscoll, of the Voice of America, Cairo, all contributed mightily to the project. Particular thanks are due Prof. Richard Yorkey (AUB), Prof. Raja T. Nasr (BCW), Dean Jack Nesmith (AUC), and Prof. Albert Khalafala (LU) for their assistance in getting test subjects. Prof. Nasr went so far as to dragoon them from a tennis court! Student subjects who were extremely helpful were Miss Mona Saba and Mr. Muhammad Ibrahim, both of AUB, who subjected themselves to more than fifty tests over a six week period. The author is grateful beyond words to these and all the other volunteer subjects who gave of their time from busy schedules. Finally, acknowledgement must be made to Mr. William R. Babcock, Department of Languages and Linguistics, University of Rochester, who must be counted as one of the best assistants, students, and tape-editors any researcher could have. All figures were prepared by him.

PREFACE

This investigation is an attempt to assess the role of the second formant transitions in cueing discrimination of velarized and non-velarized consonant pairs in spoken Lebanese Arabic. The velarized members are commonly called the emphatic consonants, though they do not include all of the consonants sometimes called emphatic. The project was sparked by an article published in Language 32:3, 446-452, by Charles A. Ferguson, "The emphatic / in Arabic", and by questions raised but necessarily left unanswered at the 1956 Summer Institute of Linguistics at the University of Michigan. Further questions were raised by Jakobson, Fant and Halle,1 and in a number of articles from the Haskins Laboratories. The course of the study involved work with informants in Easton, Pennsylvania and in Rochester, New York, synthesis at the Haskins Laboratores, New York, and further testing in the field, in Beirut and Cairo. Final tests were constructed in the Phonetics Laboratory, Department of Languages and Linguistics, University of Rochester. Later stages of the project involved the testing of acoustical features other than the second formant, though the present report is basically limited to discussion of that feature. Results of the later work will appear. Thus, the present work is largely limited to discovering the behavior of the second formant in cueing the Arabic velarized/non-velarized distinction, and to specifying some perceptual boundaries as revealed by synthetic speech study of the results of second formant variations. In this connection it is worth mentioning that the use of synthesized speech in one form or another seems to be the only practical way of exploring in a tightly controlled and economical way the effect of a given formant. Analysis of spectrograms is difficult, time-consuming, and full of inherent deficiencies, and the location, selection and matching of real speech samples adequate to the task posed by the present set of problems is excessively formidable. The writer is fully aware of the defects of speech synthesizers, and of the difficulty of equating possible 'machine thresholds' with human ones, and of the questions concerning the applicability of machine-speech perception to real speech. In view of the fact that the present task would have been effectively impossible with real speech, the risks of using synthetic speech were gladly accepted. Further, it is by now quite clear that experimentation with synthetic speech 1

R. Jakobson, G. Fant, M. Halle, Preliminaries to Speech Analysis, (= Technical Report No. 13), Second Printing (Cambridge, Mass., Massachusetts Institute of Technology, 1952), 50,

8

PREFACE

has been enormously revealing and suggestive, having made possible in a reasonable time results and types of investigations which would either have gone undone and undiscovered, or which would have been delayed by many years in the absence of synthesizers. No one suggests that a synthesizer is equivalent to a native speaker, nor that data developed in connection with synthetic speech tests are necessarily perfectly correlatable with the actual (whatever that may mean) internal language organization of a human speaker. Yet it is quite clear that linguistics and speech research would be much the poorer for the lack of the research based on synthetic speech or incorporating its results and considerations. The present study apparently represents the first large-scale attempt to take laboratory-made synthetic speech perception tests into the field; that is, to introduce acoustic phonetics into linguistic field work, as well as one of the first sizeable efforts to perform this kind of testing in a non-Indo-European language. Results to be reported elsewhere also have a bearing on distinctive features theory and on dialectology.

TABLE OF CONTENTS

Acknowledgements

5

Preface

7

I. Background and Design of the Experiments II. The Consonants and Velarization III. The Phonemes, Acoustically Characterized

11 19 24

A. Consonants

24

B. Vowels

28

IV. Test Results and Interpretations

30

A. Stops

31

B. Fricatives

34

C. Nasals

36

D. Laterals

37

E. Trills

38

V. Conclusions

39

Bibliography

44

Figures

47

I BACKGROUND AND DESIGN OF THE EXPERIMENTS

Various projects at the Haskins Laboratories and elsewhere have dealt with the perceptual attributes of different portions of the speech signal. These have concerned primarily the formant structure, including the frequency specification, the rate and direction of change, and the like. To the relatively rapid and characteristic changes lying at the presumed 'boundary' between consonant and vowel, the name TRANSITIONS has been given.1 It has been shown that these vocalic portions carry a great deal of the information leading to listener identification of the adjacent consonant. 2 The earlier work of Potter, Kopp and Green 3 identified the characteristic behavior of the vowel formants in the vicinity of specified consonants, as part of the effort to enable persons to read speech directly from a spectrographic presentation. The work at the Haskins Laboratories and elsewhere has added a considereable degree of precise knowledge of this behavior and its effects, and the work has built up a fund of recognized available techniques. In general, then, the procedure to be followed in the present study was clear. By its nature, this study involved a number of unusual problems and difficulties as well as unusual opportunities. For example, most such studies examine an extremely limited set of phonemic oppositions, covering a small range of phonetic categories, or a single category, while the nature of the emphatic/non-emphatic opposition required the present work to span the classes of stops, sibilants, nasals, laterals and flaps. Although there is sometimes a modest physical separation between the laboratory involved in the preparation of such tests as these and the test subjects, perception experiments employing altered natural or completely synthetic speech are normally essentially a laboratory-based exercise, involving easy access to necessary equipment during the course of experimentation, so that progressive refinement of the tests is possible. The present study involved an apparently unique effort to take perception testing into the field, involving in this case a separation between testing ground and 1

Alvin M. Liberman, Pierre C. Delattre, Franklin S. Cooper, and Louis J. Gerstman, "The role of consonant-vowel transitions in the perception of the stop and nasal consonants", Psychological Monographs: General and Applied, 68, 8. 1 Liberman, Delattre and Cooper, "The role of selected stimulus-variables in the perception of the unvoiced stop consonants", The American Journal of Psychology, Vol. LXV (October, 1952), 497-516. And others. " Potter, Kopp and Green Visible Speech (New York, D. Van Nostrand and Company, 1947).

12

BACKGROUND AND DESIGN OF THE EXPERIMENTS

laboratory of about 9000 miles. Thus refinement during the crucial period of the project was impossible, and failure for any cause required the abandonment of any particular test. Simple mechanical difficulties could and did influence the outcome of the project as they could never have done in the laboratory environment. In American tests involving English the acquisition of sufficient test subjects is very simple. Tests involving Arabic, however, render this matter extremely difficult, even in New York City, especially in view of requirements concerning specific nationality, recency of residence in the native culture, and the like. The present study ultimately employed 115 test subjects, tested in four locations in two nations, and representing seven nationalities in the Arabic speaking world. All of them were current residents of an Arab culture. A typical study reported in the literature involves, in addition to the various pretest forms, a normal maximum of two or three different tests. By virtue of all the questions, phonetic classes, and logistic considerations involved, the project reported here required the manufacture of something over fifty distinct tests, beyond developmental forms. Further, extensive pretesting, which is simple in the presence of the ultimate test population, could in the present case only be performed in a limited manner, thus leading to the possibility of generating a good deal of useless data, or unavoidable failure to acquire maximally useful data. Follow-up testing in the same population is also essentially impossible. Persons working with discrimination tests of this sort like to see on their graphs a well-centered display indicating that their stimulus range covered the response possibilities adequately and evenly. Predicting such coverage for a remote population, even with modest pretesting, is chancy at best. The degree of such success encountered in this project suggests, aside from good luck, the notion that a few carefully selected subjects may be just as valuable as a much larger number of more naive subjects, and that residence in another country does not unduly influence the judgements obtained from them. PROCEDURE

First steps, after exploratory work with informants, involved the preparation of tapes of real speech.4 Spectrograms were made of appropriate items, and were studied and measured. The nature and extent of formant 2 behavior were noted, and an attempt made to locate other probable powerful cues. It soon became apparent that F2 variations were the most powerful cue factor, and the obvious items to select for the study. Earliest synthesis work in the project had the form of simply attempting to copy spectrograms and trying to produce intelligible output. 5 Partly to gain further experience in synthesizing and partly to train the informant in listening to synthetic All recording work, including testing playback, was done with Magnecord PT6 and Ampex 601 machines, which are of recognized high quality, and high fidelity speakers. 5 See again fn. 1, p. 11.

4

BACKGROUND AND DESIGN OF THE EXPERIMENTS

13

speech, tapes were made from some of these early efforts, and in some cases made into small perception tests. In many cases actual word pairs were synthesized, reduced to one variable, F2 onset frequency, but attention soon shifted to single syllables, first with various vowels, later standardized. Though in the cases of a great many of the consonant pairs to be tested truly minimal pairs were found, so that testing could have been carried out with actual words, it was felt that the syllable approach was more valid and more nearly immune to accidentally introduced bias. Bias would certainly have appeared in any instance where the simplification of the pattern inherent in synthetic speech produced by chance a more satisfactory synthesis of one member of the pair than the other. In addition, the use of real items was distracting to informants, drawing attention to the unnatural quality of the voice and the generally unsatisfactory quality of synthetic speech in contrast to real speech. Moreover, the time required to paint complete word patterns is excessive unless significant additional benefit is anticipated. Once the general course of the study had become apparent, and the experimenter's skill in synthesis judged adequate, the specific procedure was to select a velarized/nonvelarized pair and to attempt synthesis of a clear repetition of each member of the pair, regularized and identical save for F2 onset frequency. When the results had been rated satisfactory by Mr. Barakat, the task became essentially one of filling in at regular intervals between the actual or posited extremes by drawing the intermediate patterns. Thus a continuous-scale presentation of five or more variants was created, all with regularized patterns identical in all respects save F2 transitions. These normally led from an unmistakably velarized phone at one extreme to an unmistakably non-velarized phone at the other, in constant intervals. With a number assigned to each pattern these were then recorded and arranged on tape into a standard discrimination test. The test was then administered to informants for their assignment of the test items into the velarized or into the non-velarized category. All perception tests were presented to the informants in one of several random orders.6 In all the work to date seeking to establish relationships between acoustic cues and articulation, a rather complete selection of the common phonetic dimensions has been covered. The voiced stops seem to have received the most attention, while voiceless stops, fricatives' and nasals have also been the subjects of some studies. In addition, some work has been done with English /w, r, 1, y/. 8,9 It has been characteristic of the papers issuing from most such experiments that they have dealt with a limited range or oppositions at a time, e.g., voiced versus voiceless stops, labial versus alveolar • These randomizations, derived from standard tables, were kindly provided by Dr. Howard Gallup, of the Department of Psychology, Lafayette College. ' Katherine S. Harris, "Cues for the discrimination of American English fricatives in spoken syllables", Language and Speech, 1: Part 1 (January-March 1958), 1-7. See also G. W. Hughes and M. Halle, "Spectral properties of fricative consonants", JASA, 28, 303-310. 8 J. D. O'Connor, Gerstman, Liberman, Delattre and Cooper, "Acoustic cues for the perception of initial /w, j, r, 1/ in English", Word, "13:1 (April, 1957), 24-43. * Leigh Lisker, "Minimal cues for separating /w, r, 1, y/ in intervocalic position", Word, 13:2 (August, 1957), 256-267.

14

BACKGROUND AND DESIGN OF THE EXPERIMENTS

versus velar place of articulation, and the like. These papers fortunately showed the way for experiments with the Arabic velarized/non-velarized consonants. But unfortunately for the experimenter, while the velarized/non-velarized opposition is itself binary, the contrast is found in all the above phonetic categories. Thus while each experiment was limited as a rule to but one contrast, a design had to be worked out and synthesizing experience gained with several manners of articulation intersecting with several places. The proposal was not simply to show that some form of F2 alteration could produce velarization discrimination between, say /I/ and /}/, and that some perhaps different F2 change would cue discrimination between velarized and non-velarized stops, but that the discrimination could be obtained in all these classes by the identical F2 feature. 10 Every eifort was made to develop a family of tests having a high degree of similarity, all or nearly all concerned exclusively with variations of F2 transitions. The tests differ otherwise only to the extent made necessary by the synthesis requirements for dissimilar phonetic entities. In a few cases minor adjustments were necessary to secure informant acceptance. In a few others it was necessary to place the consonant intervocalically rather than initially, as in the /r/ and /I/ experiments. Concerning /r/, synthesis proved inordinately difficult in initial position, and while it was relatively simple in final position, it was felt that the intervocalic one was better and more similar to the other tests. The difficulty with /I/ is apparently that /}/ is excessively rare other than medially. It was possible to synthesize initial /}/ in a manner quite satisfying to the experimenter, but one of two informants consulted reacted to it with distress, the other not at all. Other special problems and features of the work with /r/ and /1/ will be discussed later. The pattern playback spectrum is composed of 'channels', in increments of 120 cps from the fundamental frequency of 120 cps. This means that pattern variations must differ by a minimum of 120 cps in, say, frequency of F2 onset. This limitation of the machine naturally had an unavoidable effect on the experiment design, including the determination of the number of stimuli needed for each test. The frequency area normally being investigated for each contrast did not exceed approximately 600 cps. F2 onsets for /d/ and /d/, for example, average about 1800 cps and 1200 cps respectively. Thus six stimulus patterns were as a rule sufficient to move from essentially 100% velarized identifications to 100% non-velarized identifications. The discrimination functions obtained suggest that gradations finer than 120 cps would be of slight value. Secondary features which had to be determined included length of F2 transition segments, optimum length for complete stimulus patterns, and the proper handling of F3. 10 Similarly troublesome has been the fact that unlike those working with their own language, relatively little in the way of evaluation of trial stimuli could be performed by the experimenter, despite some knowledge of Arabic. All evaluations had perforce to await the next meeting with the principal informant in another city.

BACKGROUND AND DESIGN OF THE EXPERIMENTS

15

Spectrograms early disclosed that Arabic F2 transitions, particularly in the vicinity of velarized consonants, tend to be longer than English transitions. That is, there is typically a greater time span between formant onset and the reaching of a steady formant level. The experiments on English transitions, including optimum length, established a figure of 50 msec audible transition for the stops. 11 In drawing test # 1 a transition of 100 msec was used (/tii/ versus /tii/) with excellent results, but with a slight impression of over-careful speech. Transitions of only 50 msec were tried, but with sometimes poor results, the excursion of F2 in the velarized stimuli being so great that the angle of the transition was excessive. The result is an unnatural onset which seems at once over-loud and indistinct. Informant reaction was used to arrive at an F2 transition duration of 70 msec, a duration which gives the impression of a normal rate of articulation. This observation would tend to corroborate the statement 12 that the relatively standard time interval (in English) between the formant locus and the steady state corresponds to some particular rate of articulation. 18 The increased time required with Arabic stops is no doubt accounted for by the additional opposition among the stops and the increased articulatory adjustment required in velarization. In working with one-syllable stimuli the vocalic portion of the stimulus must be of reasonable length — presumably of realistic isolated-syllable length — in order to gain acceptance from the informants. In practice this meant a limitation to the long vowels of Arabic, as a stimulus with the vocalic quality of a short vowel and the length required for a CV stimulus pattern was very unrealistic. A truly short vowel, while possibly long enough for testing, would be composed entirely of a moving transition, which in the absence of a closing consonant is also unrealistic. In addition, in real speech all the short vowels tend, at least, to be distinguishable as to velarization only by their occurrence as a brief, very nearly steady-state vowel either entirely within or entirely without the velarized zone. Spectrographs evidence was used to select an optimum value of 300 msec for the entire stimulus pattern. The transitions of F1 were normally decided by the requirements of synthesis, but F3 had to be disposed of in some uniform way not prejudicial to the results. At times F3 shows a slight rise in the vicinity of velarization, but even this slight rise is often not discernible. Therefore, it was decided to make F3 absolutely flat and of uniform frequency for all experiments. Whenever possible, each stimulus pattern was presented ten times, and the minimum number of occurrences in a test was five. In accord with standard practice at the Haskins Laboratories, each stimulus was repeated once after a short interval, with a longer interval occurring before the next item. In each case the unavoidable machine 11

Delattre, Liberman and Cooper, "Acoustic loci and transitional cues for consonants", JASA, 27:4 (July, 1955), 769-773. See especially 772 " Ibid. 18 Ibid., see 769 for a discussion of the locus concept. Note that acceptance or rejection of the concept of locus is irrelevant to the observations here reported.

16

BACKGROUND AND DESIGN OF THE EXPERIMENTS

noise was begun slightly in advance of the pattern to avoid the distortion of simultaneous onset. One major decision concerning the experiment design was made rather late in the project, the necessity for it having been demonstrated by the earlier work. This had to do with the adoption of a standardized, non-occurring vowel as the frame for the consonants being tested. The course of the project was generally from the specific towards the more general, beginning with synthesis of real words and moving through regularized real word patterns to syllables. The use of a possible but non-occurring vowel is regarded as only one more step. The principal difficulty encountered in testing perception of the consonantal velarized/non-velarized contrast has of course been that of separating the velarization factor of the consonants from concomitant vowel allophones. In other words, the Arabic speaker receives an input rich in information as to velarization or lack of velarization. This is in the form both of different F2 onset frequencies and different vowel steady-state quality. The proportions of these vary with the particular vowel and consonant involved, /aa/, for example, has been described as substantially changed throughout when in the vicinity of velarization, and /r/ has been cited as a consonant depending nearly exclusively on steady-state alteration for identification as /r/. In some cases excellent discrimination can be had by varying only F2 steady-state, so long as transitions give a natural effect. This depends primarily on the particular vowel chosen. In the instance of the dental stops before /ii/, however, results are more striking with transitional variations than with steady-state variations. An opposing example was found in working with an open and backed allophone of /aa/, of the sort which normally accompanies velarized consonants. This vowel could not be made to yield non-velarized identifications, no matter how extensive the transition into the nonvelarized range. Similarly, a frame of an /aa/ allophone clearly of a sort found with non-velarized consonants failed to elicit velarized identifications despite F2 onsets very clearly in the velarized range. Transitions were progressively lowered until the stimulus ceased to be accepted as speech, still with no velarized identifications having appeared. Continuing in this line, the initial portion of some of the extremely low F2 onsets was removed, or cut back, in an effort to gain the impression of the lower onset frequency while preserving naturalness, but this failed to better the response significantly. Nonetheless, since the informant was dissatisfied with the quality of stimuli having a constant F2 onset and varying only in vowel steady-state, it could not be said that the F2 transition was without effect in determining the /aa/ contrast. But one concludes that with the allophones of /aa/, in contrast to the situation with /ii/, the transition is a less powerful cue. It was concluded, partially as the result of tests # 10, # 11, and # 1 2 , that the only way to give unity to the study and remove the bias supplied by the vowel segment was to choose a vowel as natural as possible from the standpoint of velarization and adopt it as a standard for a test series. An average vowel was therefore selected,

BACKGROUND AND DESIGN OF THE EXPERIMENTS

17

one which gives the effect of a slightly centralized mid front unrounded vowel, and which does not occur in the dialects in question except occasionally when short and in unstressed position. In the test series it is long and of course stressed. The precise acoustic nature of this neutral vowel is as follows: F1: three channels in width, center channel frequency being 480 cps. F2: three channels in width, center channel frequency being 1560 cps. F3: two low-intensity bands of energy within the channels at the frequencies of 2880 and 3000.14 The term CHANNEL, in this connection, refers to the previously mentioned channels of the pattern playback. The above description means, for example, that F1 is a solidly painted band covering the three channels corresponding to 360 cps, 480 cps, and 600 cps, and that in the resulting signal there will be energy at each of these frequencies. These correspond to three reinforced harmonics of the fundamental tone of 120 cps. This neutral vowel was used as the frame in tests # 14, # 15, # 16, # 17, # 18, # 19, # 22, # 23, and # 24. The probing conducted in tests # 10, # 1 1 , and # 12 showed that with a not excessively specialized or biasing vowel the informant could indeed divide the spectrum into appropriate successions of velarized and non-velarized consonants on the basis of F2 transitions alone. As would be anticipated, results were less clean when the subject received less than the normal amount of information. Subjects also commented that neutral vowel tests were more difficult, but it remains true that their performance was generally satisfactory. It is fortunate that the neutral vowel selected for further work provided maximal flexibility in the more or less realistic manipulation of F2 transitions. That is, formant 2 of this vowel is conveniently located at about the mid range of normally occurring second formants, and thus opens a large percentage of the lower speech spectrum to investigation without undue loss of realism. Once this vowel was standardized, a group of highly similar tests was developed. Variations other than the one being tested were internal to each test and were dictated only by the type of consonant being synthesized. The methodology employed in this study, and the philosophy of the project in general, reflect something of the history of both the author and the project. Further, some inherent limitations of the study must be understood in terms of the nature of the particular task at hand, including the gathering of data in the field. Speaking in general, many different types of work and experiments could have been carried on with respect to the problem, but not necessarily under the conditions governing this study. Some results, as well as some efforts which were NOT made, are directly related to deliberate decisions concerning what to include and what to exclude. These decisions had to be made at the beginning of the work, and abided by, in some cases, to the end 14 Full width F3 patterns give too much power at high frequencies, hence intensity is often reduced through use of thin lines for that formant. See sample stimulus patterns accompanying test descriptions in Chapter IV.

18

BACKGROUND AND DESIGN OF THE EXPERIMENTS

of the project, in the full knowledge that at least some of the decisions would turn out to be wrong. Unfortunately, in the work of this particular nature, there is little one can do in the way of changing his mind when nearly nine thousand miles from his laboratory. The writer generally works on the awesomely documented principle, ascribed to many originators, that whatever can go wrong, will. Thus a great deal of attention was given to planning an essentially simple experimental approach most likely to produce useful results under rather uncontrolled conditions, and to assuring during the field data-gathering stage the presence (by taking it along) of every conceivable piece of necessary equipment, spare parts, transformers, and the like, such as to make the project maximally independent of local conditions, and minimally subject to the vagaries of local electrical systems. The one deficiency encountered in this regard was admirably made up through the kind offices of the Voice of America, in Cairo, which provided what was probably the only example of the missing part to be found in Egypt. When one takes a project into the field, in distant lands, the phrase "Back to the drawing board" becomes entirely academic. All the work in the entire project, which ultimately studied more factors than the second formant, was based on spectrographic studies of Arabic as spoken by Lebanese and Egyptian informants located in the United States. Original efforts at synthesis of Arabic, with particular attention to the behavior of the second formant in relation to emphasis, were associated with the writer's dissertation. 15 The choice of the matter of emphasis as a subject for study indicated a need for simplicity, since the phenomenon extends over a number of phonetic classes, thus raising many more problems of synthesis, rules for synthesis, and testing procedures than are normally encountered in a single set of experiments in this field.

14

Dean H. Obrecht, Effects of the Second Formant in the Perception of Velarization in Lebanese Arabic, Dissertation, University of Pennsylvania (1961).

II THE CONSONANTS AND VELARIZATION

The traditional name for a group of Arabic consonants, including but not limited to the velarized ones, is 'the emphatics'. The Arab grammarians describe the consonants involved in "such terms as 'dark, fat, thick, corpulent, heavy' applying the Arabic word mufaxxama, 'heavy'. One of the writer's informants habitually characterizes them by striking an oratorical pose and pronouncing in sepulchral tones, "Big voice". There appears to be some disagreement as to just what constitutes 'emphasis', and which consonants are to be classed with the emphatics, though this disagreement is in matters of detail. The difficulty is in the striking variety of consonant types subsumed under the term. It is simple to detect and agree upon a component of pharyngeal-area constriction in the case of a dental, and quite another thing to agree upon the status of a consonant completely articulated in the pharynx, which may be " m u f a x x a m a par nature". 2 (/p/) /b/ IV /ml

m m m (/m/)

(/q/)

/?/

w

/H /

N

hi

/V

N /d/

N

N

141

m

N

M

Is/

M M

/?/

m

N M

III /r/

hi

M

M

Fig. 1. Consonant Phonemes of Modern Colloquial Lebanese Arabic.

Items in Fig. I 3 enclosed in ( ) either occur only in loan words or are otherwise of limited occurrence or of dubious status. Such of these as may be pertinent will be discussed later. Mention is made of the fact that although these tests were originally designed with Lebanese Arabic in mind, they proved generally quite applicable to Egyptian subjects as well. 1 Roman Jakobson, "Mufaxxama — The 'emphatic' phonemes in Arabic", in Studies Presented to Joshua Whatmough, Ernst Pulgram, ed. (The Hague, Mouton and Co., 1957), 107. 2 Jakobson, op. tit., 112. 3 The phonemic transcription used throughout this paper is that of the Department of Near Eastern Studies of the University of Michigan.

20

THE CONSONANTS AND VELARIZATION

Those phonemes always or sometimes referred to as emphatic are: /b/, /m/, /t/, 14/, hi, /?/, M, l\l, Ixl, M, hi, /q/, /H/, and /