215

Remarks on the Nature of Complete Consonantal Assimilations

James W. Hutcheson Ohio State University

0. Introduction The purpose of this paper is to clarify the nature of complete consonantal assimilations. By complete consonantal assimilations I mean those processes which convert a sequence C1C2 to C2C2 (complete regressive ), CiCi (complete progressive assimilation), or C3C3 (complete coalescence). Correspondences which illustrate such assimilations are cited from both historical change and synchronic derivations, the latter including casual speech , derivational and inflectional processes. The nature of the data makes it necessary to distinguish care- fully between correspondences of either historical or synchronic derivations which arise from a single process of complete assimilation and those which arise in some other fashion. Of the latter, two different situations obtain. Either the complete assimilations arise fortuitously as a subset of some larger, single feature, assimilatory process, e.g. a voicing assimilation rule; or they are the result of a number of processes whose cumulative effect is a complete assimilation. Such processes may be all assimilatory in nature or they may include assimilatory and non-assimilatory processes.

1. Fortuitous complete assimilations: English casual speech sandhil In two notes C.-J. Bailey (1969, 1970) has offered evidence that consonant clusters of alveolars followed by nonalveolars appear to be marked sequences and for this reason are unstable, showing a tendency to become unmarked. Such marked clusters can be eliminated in a variety of ways. Bailey cites unmarking by metathesis in the case of ancient Greek and by assimilation in contemporary English. It is the latter case that is of interest here, because the assimi- lations which do occur--and, perhaps more important, those which do not--offer important insights into the nature of complete consonantal assimilations. Bailey (1970) cites the occurrence of the following assimila- tions in rapid casual speech: righp poor, goob-bye, leab balloon, righk corner, bag guess for right poor, good-bye, lead balloon, right corner, and bad guess respectively.' However, such forms as keet track (for keep track) and bat track (for back track) do not occur. These forms are sufficient for Bailey to make his point about the marked nature of the alveolar plus nonalveolar sequence. More information of theoretical interest and importance can be obtained from this tendency toward assimilation in rapid speech. Note that forms cited by Bailey involve clusters in which the two segments share all features except the position of articulation feature alveolar. Thus t+p, d+b, t+k, and d+E assimilate completely to pp, bb, kk, and EL respectively. If, however, clusters of alveolars plus 216 nonalveolars which are not identical in all other respects are considered, no such complete assimilation occurs. Thus, right bill, bad police, right goal, and bad kid, become righp bill, bab_police, righk goal and big kid preserving the difference in voicing. These facts have been noted by Gimson (1960, 1970). Both Bailey and Gimson point out that the alveolar sibilant s will assimilate to a following palatal, e.g. horse shoe and is sure become Chorffuw] and EI3for]; however, s does not assimilate to a following labial or velar. Note that any change in the position of the sibilants would require changing more than just the position of articulation; the distinctive sibilance could not be maintained. The assimilation of alveolar n to the position of a following nonalveolar is an instance of a much more pervasive tendency of a nasal to assimilate to the position of a following segment. More- over, of all nasals, the alveolar nasal n is the most unstable with respect to position, a fact which seems to support Bailey's claim about markedness. The alveolar lateral 1 does not assimilate at all in casual speech. This is not surprising in view of the examples cited above. The assimilatory process under discussion is one which affects the feature makeup of a segment only minimally. In order for a lateral 1 to change position (other than the light-dark alternation), it must undergo a change not only in position, but in manner features as well. These observations about one kind of putative complete assimilation are offered as supporting evidence for a more general principle which claims that complete assimilations normally occur only when the segments involved are already very similar (indeed the casual speech assimilation I have been discussing is really a total assimilation of position affecting alveolar segments and is an example of a complete assimilation in the more general sense only fortuitously in the small set of instances where the segments involved are already nearly identical).

1.1. Fortuitous complete assimilations: Yakut plural inflection In Yakut, a Turkic of Siberia, nouns form plurals by a process of suffixation (Krueger 1963:74-5). The plural suffix is mlAr (where A represents a subject to regular harmonic alternations which are of no significance to the point in question). The initial segment of this plural ending remains 1 only when suffixed to a stem ending in a vowel, a , or 1 itself. When suffixed to other stems, the 1 of the plural regularly assimilates in the following way: 217

a. 1 4 t C where C is a voiceless ; there are no stem-final voiced

at 'horse' attar 'horses' balik 'fish' baliktar 'fishes' iuox 'what thing' tuoxtar 'what things' iskaap 'cabinet' iskaaptar 'cabinets' muos 'horn' muostas 'horns'

b. 1 4 d/C where C is r or y

ubay 'elder ubaydar 'elder brothers' brother' atiir 'stallion' atiirdar 'stallions'

c. 1-' n/C where C is a nasal

suoryan 'blanket' suoryannar 'blankets' xatio 'birch', xationar 'birches' olom 'ford' olomnor 'fords'

Notice that the 1 assimilates with respect to a single feature in each case. In no instance is the alveolar point of articulation lost. As was the case in the English examples the only complete assimilations are in those cases where the neighboring segment was already minimally different. Elsewhere the assimilation is partial.

2. Combinatory processes: Finnish gradation A well-known aspect of Finnish is the alternation of weak and strong stems known as Fradation, which affects - initial stops in closed . Of particular interest are cases where gradation involves a complete assimilation of stops with preceding nasals or liquids; assimilation is complete only when certain phono- logical conditions are met (see Harms 1964, Karttunen 1970). Finnish has underlying consonant clusters of the following kind: ER, nt, nk, rt, rk, lt, lk. One regular (and obligatory) process in the language is the assimilation of n to the position of a following velar, giving a derived cluster Dh. When the clusters meet the conditions for gradation (e.g. in nouns in the genitive singular) the following assimilations result:

mp mm nt 4 nn Ok 4. 00 (spelled pE).

lt 4. 11 rt rr But: lp 4 lv lk 4 1 rp rv rk 4 r 218

The assimilations involving nasals are complete in all features. Note that all examples of clusters with nasals involve a nasal and a homorganic stop, a fact assured by the operation of the nasal assimilation rule. When the cluster involves a liquid (r or 1) plus a stop, only the homorganic clusters assimilate. Such assimilations provide additional support for the claim that complete assimilations arise when the segments involved are minimally different. In the case of the nasal cluster nk the segments become increasingly more alike through the combination of two individual assimilation rules.

2.1. Combinatory processes: Greek sonorant clusters Some combinatory processes which result in the complete assimilation of two consonants involve more than just a succession of single feature assimilation rules, as is shown by developments in ancient Greek. The Indo-European clusters *VRs and *110 become VR in most Greek dialects; however, the development of these clusters in Lesbian and Thessalian is to VRR. Kiparsky (1967) has offered an explanation of these developments which, he claims, shows that they are a subset of a larger unitary process involving resonant and glide clusters. Although I do not agree fully with his analysis, the part of that analysis which is of present interest seems well motivated. By simply observing the historical correspondences it would seem that, in the case of Lesbian and Thessalian, there are two kinds of assimilation at work, a progressive assimilation of *VRs to VRR and a regressive assimilation of *VsR to V. Kiparsky has shown, however, the a rule changing s to h is needed anyway, and so the clusters which assimilate are VRh and VhR. There is, more- over, other evidence which leads Kiparsky to posit a rule of meta- thesis which converts a cluster ya to laa. Ultimately then the clusters which undergo differential treatment in the dialects have been reduced to one cluster VhR. Notice that such a solution makes reasonable a development of Cal from an original *VRs cluster, a virtue which no solution not involving metathesis shares. To the rules which Kiparsky has adduced I would add a rule voicing h in the environment V R. Thus the change involving the two Greek dialect groups involves the following rules:

a. s becomes h in a sonorant environment b. h metathesis c. h voicing d. D assimilation to a preceding vowel - regressive assimilation of h to a following resonant.

This combination of processes resulting in the ultimate complete assimilation of the Indo-European clusters *VRs and *VsR to VRR in 219 Lesbian and Thessalian illustrate very well the second type of combinatory assimilation. The fascinating question of why the dialect split developed, i.e. why one dialect group shows assimilation of to a resonant and the others show assimilation to a preceding vowel, is of course not answered by this analysis. On the other hand, if this analysis is correct we at least know how to ask the question properly.

3. Single process complete assimilation: the Arabic definite article The dialects of Arabic spoken in Syria (Cowell 1964), Morocco (Harrell 1972), Iraq (Erwin 1963) and the San:di dialect of Egypt (Khalafalla 1969) show complete assimilation of the 1 of the definite article when that article is prefixed to a noun which has an initial dental, alveolar, or . Elsewhere the 1 remains unassimilated. For example, from Iraqi (Erwin 214-5):

a. Unassimilated

?akil 'food' 1-?akil 'the food' beet 'house' 1-beet 'the house' fikra 'idea' 1-fikra 'the idea' qisim 'part' 1-qisim 'the part' kaatib 'clerk' 1-kaatib 'the clerk'

(Similarly words with initial x, y, 11, T, m, h, w, y, p, g.)

b. Assimilated

timman 'rice' t-timman 'the rice' Ooob 'shirt' 0-0oob 'the shirt' junta 'suitcase' j-junta 'the suitcase' diinaar 'dinar' d-dinar 'the dinar' 6ibbaan 'flies' 6-6ibbaan 'the flies' rukkaab 'passengers' r-rukkaab 'the passengers' zibid 'brother' z-zibid 'the brother' s-sana 'the year' .,sana 'year' ... - sahar 'month' s-sahar 'the month' suura 'picture' s-suura 'the picture' 6aabuV 'officer' 6-6aabut 'the officer' Vooba 'ball' t-tooba 'the ball' leela 'night' 1-leele 'the night' naar „ 'fire' n-naar , 'the fire' caakuuc 'hammer' c-caakuuc 'the hammer'

In the colloquial Arabic spoken in Cairo the assimilation extends optionally to initial k and & (Mitchell, 47n: Abdoul-Fetouh 1969). The assimilations in all dialects are complete. Of particular importance, however, is the fact that in each dialect, including Cairo Arabic, the assimilations take place between similar sounds and extend in the case of Cairo Arabic hierarchically. Thus 220

we would not expect ?fl.-k to assimilate to ?ik-k unless ?il-t assimilated to ?it-t also. It is of further interest that the extension of the assimilatory process to include velar stcps is optional, showing the customary tendency of a process to operate optionally at first as it extends its domain. Both the Arabic examples just cited and the Yakut assimilations treated in 1.1 show one feature that is of further interest, in that they offer evidence that not all assimilations are determined by purely phonological conditions. In both cases the segment which is assimilated is a part of a grammatical affix. The weakness of the 1 which allows it to be assimilated is determined in both by the facts having to do with grammar. This is particularly easy to see in the case of Arabic, where there are many clusters exactly like those treated above but which do not assimilate. These facts are of course not surprising, as nonphono- logical conditions have been seen to constrain many sorts of phonological processes (see Zwicky 1970, 1972a). However, once the assimilation process has been triggered by (whatever) grammatical factors, it obeys constraints that are purely phonological, e.g. the constraints involving minimal difference and the hierarchical extensions of the process. Such evidence as that just cited should serve, however, as a warning to use extreme caution when attempting to define notions of weakness and strength on purely phonological grounds, as is done, for example, by Grammont.

4. Latin assimilations in derivations I now consider briefly the phonological behavior of the Latin prefixes wiz, con-, and ab- (VaNngnen 1963:62-8 for Vulgar Latin; Buck and Hale 1966:24-5 for the classical language). Con- shows regular assimilation of the n to the position of a following segment. Before 1 and r, the n tends to assimilate completely, as in cor-rump5, car-rip-1:6, and col-lig6. However, there are a number of combinations of con + liquid; especially con + 1, where the assimilation does not take place. Note that in the assimilation of con + 1, r to coll and corr, the assimilation is to the manner of a more sonorous segment articulated at the same point. The d of the prefix ad- shows a tendency to assimilate to nearly every following consonant; the tendency seems strongest when the following consonant is similar in position or manner. Thus, particularly in Vulgar Latin, examples of ad + t, r, n, 1, s, k, E. becoming att-, arr-, ann-, all-, ass-, akk-, and agg- are common, while the assimilation of ad-m to amm- is less regular (ad + _IRE seems quite regular).3 The Latin prefix ab- shows no such ready assimilability. Indeed where clusters would arise that might be difficult, the prefix takes the form abs or H. There are many details of the assimilations (or lack of them) that await a further, much more detailed analysis of the data from Vulgar Latin and the development of the Romance dialects, especially Italian; I expect corroborative evidence for the hierarchical nature of constraints on complete assimilations. 221

5. Conclusion: the principle of similarity In this paper I have attempted to show that complete consonantal assimilations arise in a variety of ways, only one of which involves a single process which completely assimilates all features of one segment to those of an adjacent segment. However, in each type of assimilation one fact remained constant. Wherever unlike segments of consonants are converted to like sequences, the rule which finally accomplishes this assimilation operates on a sequence of segments which differ very little in feature makeup. In the case of genuine single-process complete assimilations this requirement (that the segment be similar in feature content) is extended in such a way as to require that the process operate along strictly hierarchical lines. That is to say that should the process be extended (as was the case in the Cairo dialect of Arabic), it will include the next most similar segment or class of segments within its domain of operation. This constraint on the operation and extension of rules of complete assimilation I have called, for want of a better name, the principle of similarity. To my knowledge this particular constraint on complete consonantal assimilations has only been mentioned once in the literature (Kenstowicz 1970 who mentions the nature of this hierarchy in a slightly different context, where he claims that complete assimilations imply partial assimilations but not vice versa). Finally, it should be noted that the constraint on the operation of complete assimilations and the hierarchical constraints on the way in which such processes may be extended are like all constraints on the operations of natural processes; they are system-external, determined by the nature of thearocess itself and by the phonetic makeup of the segments involved."

Footnotes

. 1. For a discussion of casual and fast speech phenomena and their theoretical significance see Zwicky (1972a) and Dressler (1972). 2. I use Bailey's nonce spellings. 3. I have been purposely vague about the hierarchies of these assimilations. The evidence is not clear and is often contradictory (Vglanen 1963:63). Despite this the point concerning the differ- ences between con- and ad- on the one hand and ab- on the other are still valid. 4. For further examples of works within this framework see Stampe (1972a and b) and Zwicky (1972b).

References

Abdoul-Fetouh, Hilmi M. 1969. A Morphological Study of Egyptian Coloquial Arabic. The Hague: Mouton. Bailey, C.-J. 1969. A possible explanation for an assimilation curiosity. Working Papers in Linguistics 5, University of Hawaii. . 1970. Toward specifying constraints on phonological metathesis. Linguistic Inquiry 1.3.348-50. 222 Buck, C. D., and W. G. Hale. 1966. A Latin Grammar. University of Alabama Press [reissue of 1903 ed.]. Covell, Mark. 1964. A Reference Grammar of Syrian Arabic. Georgetown University Press. Dressler, Wolfgang. 1972. Allegroregeln rechtfertigen Lentoregeln: Secondlre des Bretonischen. Innsbrucker Beitrike zur Sprachwissenschaft, Band 9. Erwin, Wallace M. 1963. A Short Reference Grammar of Iraqi Arabic. Georgetown University Press. Gimson, A. C. 1960. The instability of alveolar articulations. Le Maitre Phonetique 113.7-10. . 1970. An Introduction to the Pronunciation of English. London: Edward Arnold. Grammont, Maurice. 1965. Traite de phonetique. Librairie Delagrave, Paris [8th ed.]. Harms, Robert. 1964. Finnish Structural Sketch. Indiana University Press. Harrell, Richard S. 1962. A Short Reference Grammar of Moroccan Arabic. Georgetown University Press. Karttunen, Frances. 1972. Problems of Finnish Phonology. Ph.D. dissertation, Indiana University Kenstowicz, Michael. 1970. On the notation of in Lithuanian. PIL 3.1.73-113. Khalafallah, Abdelghany A. 1969. A Descriptive Grammar of Safi:di Egyptian Colloquial Arabic. The Hague: Mouton. Kiparsky, Paul. 1967. Sonorant clusters in Greek. Language 43.3. 619-635. Krueger, John R. 1962. Yakut Manual. Indiana University Press. Mitchell, T. F. 1962. Teach Yourself Colloquial Arabic. London: English Universities Press. Stampe, David. 1972a. The natural history of . Papers from the Eighth Regional Meeting, Chicago Linguistic Society, 578-90. . 1972b. Dissertation on natural phonology. Ph.D. dissertation, University of Chicago. Vganlinen, Veikko. 1963. Introduction au Latin Vulgaire. Paris: Klincksieck. Zwicky, Arnold M. 1970. Auxiliary reduction in English. Linguistic Inquiry 1.3.323-36. . 1972a. On casual speech. Papers from the Eighth Regional Meeting, Chicago Linguistic Society, 607-15. . 1972b. Note on a phonological hierarchy in English. In Linguistic Change and Generative Theory, ed. R. Stockwell and R. Macaulay, Indiana University Press, 275-301.