462

The Weakening Chain in Natural Phonology

Larry Nessly University of Michigan

Current theory recognizes two ways in which rules can be re- lated or grouped under a single generalization: one, by collapsing them into a single rule by means of notational conventions, that is, by formally relating them; and two, by identifying a function- al relationship among them, which involves specifying a target output that the rules conspire to produce. In this paper I will discuss two examples of a "weakening chain," that is, processes which serially combine to weaken a segment to the degree that it eventually deletes, and will consider whether these two examples are the same kind of phenomena as are handled by -the standard methods of relating rules. That they are not is suggested by the following facts. While the processes that combine to form the weakening chain have clearly related similarities, their dissimi- larities make it difficult to collapse them into a single process, and they also do not share the properties of rules that are nor- mally identified as conspiratorial.. In discussing these examples and means of describing them, I will first describe the usual means of relating rules, then describe the two examples of the weakening chain, and finally consider whether these examples lie outside the domain of the standard devices for relating rules.

1. Current Ways of Relating Rules

In discussing methods of relating rules, I will first discuss the method of formally relating them. The purpose of this method is to avoid having to repeat identical parts of similar rules. For example,rules (1) and (2) given below can be collapsed using parentheses into rule (3), in which the parts of rules (1) and (2) that are identical are stated only once.

(1) V 3 stress)/ Co# (2) V 4 L+ stress)/ Co + V# (3) V 4 C+ stress)/ Co (+ V)#

Such a collapsing device can apply only when the rules involved share identical parts and when the differences fall into a few basic permitted categories. For example, rules (1) and (4) below cannot be combined, since the environments in which the rules apply are too different.

(1) V 5 El- stresS)/ Co# (4) V stresp]/E7 long] As another example, rules (1) and (5) below cannot be combined, 463 since both the phonological change and the environment is dif- ferent.

(1) v FF stress]/ Co# (5) V * 0 / #

There is also the further requirement that the two rules have to be adjacently ordered for them to be collapsed. A general state- ment of the kinds of collapsing devices that are allowed (and the implicit conditions on stating relatedness) is given in Chomsky and Halle (1968). An advance over the standard theory of collapsing is the con- cept of a "universal rule." Examples of this are given in Chen (1971), Foley (ms), Lightner (1970), and Schourup (1972). The concept is based upon the fact that natural processes in phonology are favored in some environments more than others. As examples, palatalization is more common before higher front than before lower front vowels; nasalization is more common for lower vowels than for higher vowels; and final labials delete or merge with other consonants more than velars do. Not only is one pole of the opposition that is involved (e.g., "high" versus "low" for vowels) favored over the other one, but each gradient. between the poles is favored more as the gradients approach the most favored end of the spectrum. For example, not only are low vowels favored over high vowels for nasalization, but mid vowels are more favored than high vowels and less favored than low vowels. (Further examples and evidence can be found in the ref- erences above.) Given the gradated sequences in which natural processes are favored to apply, it is possible to write universal rules for such processes, in which the specific properties of application in any one language can be located along the hierarchical se- quence of application of the sequence universally. An example (somewhat changed for presentation purposes here) can be taken from Chen (1971), to the effect that front vowels are more likely to palatalize the higher they are. ' The example is given in (i) below. V (i) C *palatalized / highbackV language-universal constraint:ot?..• at language-specific values: m 4 Mandarin m 2 Spanish m m. 1 French

Given that high vowels are given the value 4, higher mid vowels 3, lower mid 2, and low vowels 1, plus the convention that 0 means 'does not apply,' the rule states that only high front vowels have caused palatalization in Mandarin, that high and mid front vowels have done so in Spanish, and that all front vowels 464 have done so in French. The main improvement of universal rules over standard col- lapsing devices is that universal rules can combine rules from several different languages into one single generalization, whereas standard collapsing devices only apply to rules in a single language. This ability of the universal rules to state generalizations about rules taken from a variety of languages is an important one. Having looked at several ways of collapsing rules (or re- lating them formally), we can now look at a way of relating rules on a functional basis. Rules related in this way are said to be members of a "conspiracy," and examples of conspiracies can be found in Haiman (1971, 1972), Risse§erth (1970), and Lovins (1971). In a conspiracy some target output is specified, and a variety of different rules is shown to work together to guarantee that the target output is derived. The best known example is probably that found in Kisseberth (1970), regarding the nonoc- curence of consonant Blusters at the beginning or end of in Yawelmani . This is usually described in such dis- junctive.terms as that words may not begin or end with consonant clusters, nor may more than two consonants occur sequentially within a word. There are a variety of devices in Yawelmani to avoid nonpermitted consonant clusters: (1) morphemes normally are allowed to begin or end with consonant clusters, but are not per- mitted to contain triconsonantal clusters; (2) when a morpheme beginning or ending in a consonant cluster is combined with another morpheme beginning or ending in a consonant, so that a triconson mental cluster results, then one of the consonants deletes; (3) another resolution of a nonpermitted consonant cluster that arises from morpheme combinations or underlying final clusters, is to insert a into the cluster to break it up; (4) still another aspect of the conspiracy is that vowel deletion cannot apply in those cases where it would derive a nonpermitted consonant clus- ter. At least four different measures (morpheme structure con- ditions, consonant deletion, vowel epenthesis, and a derivational constraint on vowel deletion) combine to guarantee that Yawelmani does not have a nonpermitted consonant cluster on the surface. This combination of a surface target, plus a variety of different measures to produce that target, is what characterizes a con- spiracy. A few more examples of.conspiracies will help to show the different forms that'this phenomenon can take. Reiman (1971) discusses a target of penultimate stress in verb forms in the Vallader dialect of Romantsch (in Switzerland). One aspect of the conspiracy to guarantee this target is that verb roots that are two syllables long and have initial stress, change the form of the root to allow penultimate stress when a clitic or desinence is added to the root, since clitics and desinences act as part of the verb form for stress purposes. Another device is that im- personal clitic pronouns have different shapes when the clitic Z65 is inverted to a position after the verb, with one shape to give penultimate stress when the form would otherwise have final stress, and another shape to preserve penultimate stress when that is the stress that the verb would otherwise have. Another device involves a rule which truncates a clitic pronoun or a verbal desinence un- der conditions that preserve penultimate stress. Still another device is that the second plural endings for the first and second conjugations in the present indicative have been replaced by the second plural endings of the imperfect indicative, in order to preserve penultimate stress when the second person pronominal clitic is deleted. (The original endings would have given final stress after the deletion of the clitic.) Here again a number of devices, in this case alternant morpheme shapes, a truncation rule, and a case of paradigmatic borrowing, conspire to produce a given surface target: penultimate stress in verb forms. Notice that the devices involved are so diverse that they could not pos- sibly be collapsed into a single notation, although they are re- lated phenomena. As another example, Lovins (1971) posits a conspiracy in- volving tones in L)mingp, a group of Bantu dialects spoken in part of the equatorial Congo. In this language there are a num- ber of tone assimilation rules that change the underlying dis- tribution of high and low tones. The posited target is that the original contour of high versus low tones in the underlying se- quence (the underlying "melody") is still preserved after all the tone assimilation, segment deletion, and reduplication rules have applied. The conspiring rules are the tone assimilation rules, which work together to assure that the underlying melody is preserved on the surface even though the original forms have been changed considerably. In comparing her example with Kisseberth's example of a conspiracy, Lovins notes the following difference (pages 475-476):

...Kisseberth's rules act independently--each is capable of achieving the target form by itself, and the result is a conspiracy only in a rather loose sense. In 1.3mingp, several rules may work together in a single conspiratorial derivation to produce a result that some of them...would not be able to produce individually.

As a final example, Haiman (1972) proposes a target in Turkish to the effect that o occurs only in the first of words. He cites three devices which conspire to produce this result: (1) with the exception of two suffixes that anomalously do not undergo , no suffixes contain the vowel o; (2) the part of vowel harmony that rounds vowels applies only to high vowels, so that underlying a's after the first syllable cannot be rounded to produce o in that position (in that vowels are analyzed as harmonizing to the first vowel in the word); and (3) with the exception of compounds, reduplicated forms, and 466

loan words, roots have o only in the first syllable. Haiman also proposes several other examples of conspiracies, one of which is especially interesting because it involves a tendency that languages are posited to follow. The tendency is that the vowel inventory reduces to i, a, and u in syllables having lesser prominence (un- stressed syllables in stress languages, and non-initial syllables in heavily suffixing languages). whether this analysis is correct, it at least goes beyond the description of a single language and talks about general linguistic tendencies. It appears that languages also have a tendency to avoid some "negative" targets (large consonant clusters, adjacent vowels, adjacent stresses), as well as to promote some positive targets (such as open syl- lables). To summarize the discussion so far, there are two major modes of capturing relations between rules. One involves formal relations, and can be subdivided into standard collapsing and universal rules. Standard collapsing makes use of a set of ab- breviatory devices which can apply when rules in a given language share identical parts, when the nonidentical parts meet a number of conditions, and when the rules are ordered next to each other. The other subdivision, universal rules, establishes a notation for describing the characteristics of natural processes within any given language in terms of permissible forms that the pro- cesses can take. This notation is based upon the fact that natural processes apply directionally across some phonological parameters (for example, from high vowels to low vowels), and that the process in a given language can be identified as having advanced to a given point along this parameter (for example, affecting high and mid vowels, but not low vowels). The second mode of relating rules concerns conspiracies, in which a target output is specified, and then a diverse set of rules apply in such a way that the target is achieved. The basis for relating these rules is their functional similarity, in that they all work to produce a given output.

2. Properties of the Weakening Chains

Having surveyed the chief ways of relating rules, I will now look at the properties of weakening chains. These involve degrees of weakening segments in certain environments, with the ultimate weakening being deletion. First I will consider an ex- ample of consonant weakening and then an example of vowel weaken- ing. Consonants tend to weaken intervocalically. It is well-known that voiceless stops become voiced intervocalically. Examples for Danish and Spanish can be found in Foley (MS: 3). Voiceless also become voiced intervocalically, as in German and Italian. Voiced stops in this position also tend to become fricatives, which is a well-known fact about Spanish, and which also occurs in Danish (see Foley, p 2). Fricatives also have a tendency to delete intervocalically, as has occurred historically L.67 in Spanish (Foley, p 2). These examples illustrate the tendency of consonants to weaken intervocalically. It is also true that spirantization occurs preferentially from velars to dentals to labials, since in North German velars have spirantized but den- tals and labials have not, (Foley, p1) and in Danish velars and dentals have spirantized but labials have not (Foley, p 2). Labials are also more resistant to deletion, since velars and dentals have deleted in Spanish and French, but labials have not (Foley, p 2). I haven't found any evidence for preferentiality in the voicing of stops or fricatives, so that it is not clear thatintervocalic voicing has the same preferentiality as spiran- tization and deletion. It could be argued that consonant weaking in intervocalic position is mainly a European phenomenon, since the examples above were all from European languages. However, Chen (1971: 7-8) presents evidence for the same processes applying inter- vocalically in Fuzhou, a Northern Min dialect in China. Voice- less obstruents in this position become fricatives if they are labial, sonorants (LLD if they are dental, and they delete if they are velars. Such evidence not only shows that these phenomena are not limited to European languages, but also shows the connection between these processes as well as the preferential weakening of consonants from velars to labials. Another possibility, which needs to be investigated, is that this consonantal weaken- ing is restricted to languages having certain other properties, independent of what part of the world they are found in. It could also be argued that what is involved here is not intervocalic weakening, but rather the assimilation of consonants to surrounding vowels. This is the traditional explanation for intervocalic voicing of consonants, and the spirantization of these consonants could be explained in terms of an assimilation to the continuancy of the vowels, as claimed in Schachter (1964: 344). deletion could be further seen as an assimilation to the nonconsonantal nature of the surrounding vowels, which leads to deletion. However, if what is involved is feature- changing, then there needs to be an explanation for the liquid in Fuzhou. Also, the explanation of deletion would have to be more carefully argued. In either case, whether these changes are to be analyzed as weakenings or assimilations, they still present a problem for theories relating rules, as we will see in the next section. Vowels also undergo weakening in a generalized environment. The processes that are relevant here are reduction, devoicing, and deletion. With respect to vowel deletion, Bell (1971: 53-54) observes that "high vowels and reduced vowels are most prone to loss," that "accented vowels (whether by length, pitch or stress) resist loss," that "loss is sometimes favored in pre- and post- stress positions," that "loss next to sonants is common," and that "between identical consonants is another favored environment." Regarding vowel devoicing, Greenberg (1969) observes that there 468

is a "preference of voiceless vowels for ... weak stress, short length, and low pitch" (p 160); that high vowels are favored with regard to voicelessness (pp 162-163); that "every voiceless vowel is either preceded or followed by silence or a voiceless plain (i.e., nonglottalic) sound" (p 164); that "a favorite position for voiceless vowels is utterance final, while occurences in utterance, or indeed in word initial are excessively rare" . (p 164); and on p 173, fn 42, he advances with supporting evidence the hypothesis that'"final vowels are more likely to be lost after unvoiced consonants." With respect to the great similarity in the conditions on vowel deletion and vowel devoicing, Greenberg (p 172) goes so far as to say the following.

The only important respect in which a real difference seems to exist between the conditions for voiceless- ness in vowels and the conditions for vowel loss is in the preference for a voiceless consonantal environment for the former which does not appear in most formulations regarding vowel less.

With respect to vowel reduction, Miller (1972) notes that reduction is favored for unstressed nontense vowels (p 482), that "when only some of the vowels of a system are reduced, the reduction often seems to proceed in an order from the low to the high vowels" (p 485), that "unstressed uncolored Cnonround, non- palatal] vowels raise" (p 485), and that "open medial syllables apparently favor reduction and closed syllables disfavor it" (p 487). With respect to the last issue of syllable closure inhibiting vowel reduction, we see the same phenomenon in English, which is most conveniently illustrated by the pair banana/bandinna. The usual explanation for this phenomenon in English-W-0747 the closed syllable receives some degree of stress. It is equally possible that the syllable closure itself directly affects the reducibility of the vowel, without any in- termediate step of assigning stress. In this section I have presented the characteristics of con- sonant and vowel weakening without commenting very much on the parallels appearing between the related subparts of each kind of weakening. In the next section I will discuss the parallels and ways of describing them.

3. Relating the Links in the Chains

The first part of this section will focus on the parallels among the subparts of consonant weakening, and also of vowel weakening, so that it is clear that there are relations that need to be captured. Intervocalic voicing, spirantization, and deletion share one major property: they all occur intervocalically. They are also all instances of weakening. Evidence that consonant voicing is a form of weakening is given in Foley (MS). We should 469 also note the parallel preference for velars over dentals over labials suggested by the evidence for spirantization and deletion. Finally, it is of importance that spirantization is generally limited to applying to voiced consonants, and that deletion nor- mally applies to fricatives. This shows the serial relationship among these processes, and suggests that each process is merely an additional extent of the overall process which increasingly weakens the consonants until they delete. Given these parallels among these processes, it is unsatisfactory to represent the processes as completely independent phenomena. With respect to these processes, it is possible, as mentioned above, that what I am calling "consonant weakening" is really "consonant assimilation to vowels." One point to be made is that even if what is occurring is consonant assimilation to vowels, the steps in that process still need to be formally related, since their similarities are surely not accidental. Also, it is peculiar that the ultimate assimilation of these consonants to the surrounding vowels should be deletion, whereas in most other cases of such assimilations the consonants become more like the neighboring vowel rather than deleting. I will not pursue this question any further here. The examples of vowel weakening also have clear similarities. Long, stressed, or high-pitch vowels are not favored for these processes. While tenseness in vowels inhibit reduction, it is not clear that it also inhibits devoicing and deletion, however. It is also likely that devoicing and deletion are favored in open syllables, as reduction is, and it is unlikely that every case of these processes being disfavored in closed syllables is to be accounted for by some kind of "half-accent." A further similarity is that reduced vowels are favored to delete, with the voiceless environment being the environment of vowel de- voicing. It is also clear that devoicing, reduction, and de- letion all involve weakening of the vowel. Besides having similarities, these examples have clear differences. Vowel devoicing does not occur between two voiced consonants (or two glottalized consonants), while it is far from clear that reduction and deletion are constrained in the same way. Also, devoicing and deletion preferentially apply to high vowels, while reduction preferentially applies to low vowels. While deletion is "sometimes" favored next to stressed syllables, it seems that reduction is often blocked next to stressed syl- lables. Deletion and devoicing are probably both common next to sonants and between identical consonants, but the same is not clearly true for reduction. While unstressed "uncolored" vowels rise as part of their reduction, voiceless vowels do not seem to rise. Other differences can also be cited. Some of the differences cited above can probably be accounted for in terms of the phonetic conditioning of the processes. For ex- ample, it is expected that vowels do not devoice in a completely voiced environment, whereas such a condition is less significant 470

for deletion and reduction. Furthermore, Miller (p 485) offers the following explanation of why low vowels preferentially reduce.

If it is the case that lack of stress is accompanied by lack of articulatory effort, and if part of this lack of effort is failure to increase jaw opening for vowels that are originally more sonorant, then a failure to make distinctions of vowel height in reduced positions does not seem extremely peculiar. Perhaps the increase in jaw opening might be a grosser movement which is more easily acquired but which, since it re- quires more effort, is more likely to be abandoned.

Low vowels are reduced first, then, because they are in greatest conflict with the posited tendency toward pronouncing unstressed vowels with less jaw opening. In contrast with reduction, de- voicing preferentially applies to high vowels. Ian Catford (personal communication) has suggested that high voiceless vowels are still audible because of the turbulence through the narrow oral channel, whereas low voiceless vowels are almost inaudible because of the lack of turbulence, and that therefore low voiceless vowels are generally avoided. It is still unclear why high vowels are preferentially deleted. In summary, then, the elements of consonant weakening and vowel weakening have too much in common with the other elements of each process to be mere coincidence. Since the purpose be- hind explicitly relating similar rules is to distinguish between relatedness and coincidence, it is necessary somehow to relate the subparts of each general process. The first, and most basic question, is whether to consider the subparts of, say, vowel weakening to be aspects of a single general process, ortobeseparate processes which have a similar function (to "weaken"). Clearly they cannot be collapsed within the standard theory of formal relations, because there are too many differences between each one. Each process could be de- scribed to some extent by a universal rule, but the three could not be collapsed into a single universal rule. Such limitations suggest the need for a new method of formally relating rules, if (as I believe) the subparts are in fact merely aspects of a single general process of vowel weakening. Furthermore, I believe that most of the differences between the subparts could be explained in terms of their phonetic motivation if we knew more about . Given adequate phonetic explanations of these phe- nomena, I predict that many of the differences would be secondary, low-level effects of a single general difference, rather than major independent differences themselves. I will discuss this again shortly. Consonant weakening could probably be handled by some modi- fication of Foley's schema for universal rules, since that schema explicitly allows hierarchies of strength, although the notation L71 would not capture what is natural about the sub-processes. As argued in Nessly (1972), to cite one paper that discusses this, to the extent that these processes are phonetically conditioned, they need to be described in terms of that conditioning. Foley's notation might succeed in terms of descriptive adequacy, but it would fail on the grounds of explanatory adequacy, since it would not explain why these processes occur. If we reject the possibility of collapsing these processes into single general processes of weakening, then we have to con- sider other ways of relating the subparts. One possibility is to consider these weakening chains to be conspiratorial. The target is specified as'being deletion, and the separate vowel weakening processes form one conspiracy, while the consonant weakening processes form another. It appears, however, that the current theory of conspiracies is not constructed to handle such a possibility. Conspiracies are currently formulated to describe facts about a single language, in which a variety of attested rules combine to efficiently produce a given target out- put. There is no provision for combining rules from separate languages that individually work to produce a given target. Japanese is generally regarded as having vowel devoicing but not vowel deletion or vowel reduction. No one would say that Japanese has a conspiracy to weaken vowels. Just as standard collapsing devices were improved upon by universal rules, conspiracies can be improved upon by allowing "universal conspiracies." A universal conspiracy is one whose target is generally sought in languages throughout the world, but which can be achieved in any one language by one of several possible means. Evidence for such a conspiracy would be different examples from different languages working toward a common end. For example, if Language A eliminated large consonant clusters only by deleting consonants, and Language B eliminated them only by inserting vowels to break them up, then together they would be evidence for a universal conspiracy. Furthermore, the theory, if it was to parallel the capacities of universal rules, would have to allow the descrip- tion of hierarchical applicabilities that have not been consum- mated. For example, if consonant deletion spreads from deleting only stops to deleting all abstruents and resonants, and from deleting only velars to deleting all consonants, and if vowel in- sertion spreads from inserting vowels only into word-internal clusters to inserting them into all clusters (word-internal and word-marginal), then the following would be valid (although in- complete) evidence for a Universal conspiracy: Language A de- letes only velar obstruents in clusters; Language B deletes velar and dental consonants in general in clusters; and Language C inserts vowels only into word-internal clusters. Given such modifications in the notion of a conspiracy, can we now describe the relations between elements in a weakening chain? I think not, although the argument has its subtleties. Before considering reasons for rejecting the universal conspir- acy analysis of weakening chains, however, it might be good to 472

look at reasons for accepting such an analysis. One, the analysis is able to relate the different subparts without trying to de- scribe them as constituting a single large process. Two, the analysis is able to accept evidence from a variety of different languages having different forms of the basic processes. Three, the analysis explicitly recognizes that the end result of the weakenings is deletion of the segments involved. There may be other advantages that I have not recognized. The problem with analyzing weakening chains in terms of universal conspiracies is that the parts of the conspiracy are too similar to each other. An optimal conspiracy makes use of , a large number of widely different means to achieve a single target output. An optimal weakening chain makes use of a serial concatenation of extremely similar processes, each of which pro- duces an output that is incrementally closer to the deletion of the segment in question. Using the term "target chain" for a "weakening" chain that involves processes other than weakening, the ideal forms of target chains and conspiracies are different, since one requires almost identical subparts while the other re- quires maximally different subparts. Furthermore, if the dif- ferences between the subparts of a weakening chain are details that are consequences of the different phonetic conditioning or dynamics of the subparts, then the subparts are actually closer to each other than the details would suggest. This raises a new point. Two rules like (6) and (7) are completely unrelated.

(6) V &back] / C+ high pitchj # (7) V .k / G- voice] 5—Toronall

This is because there is no relation between the rule change and the rule environment, so that each difference between the rules is an additional, independent difference. However, it is not correct to say that vowel reduction and devoicing are equal- ly unrelated processes because one applies to low vowels first and the other applies to high vowels first. If the explanations given earlier for these facts are correct, then the difference is a natural consequence of the basic difference between re- duction and devoicing in general, and is not a further, inde- pendent difference once the primary effects of reduction and devoicing are identified. To consider such differences inde- pendent is similar to saying that graduate students are generally more educated than undergraduates and furthermore they know more. Just as knowing more is a consequence of being more educated, so it appears that the preferential applications of reduction and devoicing are consequences of the difference between a vowel pronounced with diminished effort and a vowel that is voiceless. The two facts do not seem to be independent. Perhaps it would be possible to clarify the status of weak- ening chains in a manner suggested by Charles Pyle (personal 473 communication). Weakening chains are different from traditionally collapsed rules, universal rules, and conspiracies, and cannot be expected to be describable in terms of those analyses, since all of these phenomena have different principles of motivation. A universal rule is motivated by the differential susceptibility of sounds to phonetic processes; a conspiracy is motivated by the attempt to produce a given output; and a weakening chain is motivated (particularly in vowels) by an attempt to diminish the articulatory prominence of a sound that is given diminished structural prominence. The motivation in neither of the weakening chains is one of deleting sounds, but is rather one of reducing articulatory effort in producing nonprominent sounds. Given the nature of the input sound and the properties of different environ- ments, we can account for the observed differences that are pro- duced. The analysis, then, should not focus on elements of "weakening," but should rather focus on aspects of underlying principles which produce the effects that we can call weakening. There are several aspects of Pyle's proposal that I agree with, namely that weakening chains are different from conspiracies and universal rules, and that the different manifestations of vowel weakening, which is the more complicated case, should be accounted for as secondary phonetic phenomena. What the nature of the un- derlying principle is (which I take to be a phonetic process of weakening), is still an unsolved question, and one which needs vastly more investigation.

Note

I would like to thank David Stampe for stimulating me to look futher into this topic, by his comments on an earlier treat- ment of this (Measly 1972). I would also like to thank Richard Rhodes for suggesting the valuable term "weakening chain," and Charles Pyle for comments that improved the quality of the paper.

References

Bell, Alan (1971) Some patterns of occurrence and formation of syllable structures. Working Papers on Language Universals 6: 23-137 Stanford U. Chen, Matthew (1971) Metarules and universal constraints in phonological theory, Project on Linguistic Analysis Reports, Second Series. No. 13. U. of California, Berkeley. Chomsky, Noam and Morris Halle (1968) The Sound Pattern of English. New York: Harper and Row. Foley, James (MS) (no date) Systematic Morphophonology. Unpublished paper. Greenberg, Joseph H. (1969) Some methods of dynamic comparison in linguistics. In jaan Puhvel, ed., Substance and Struc- ture of Language, Berkeley and Los Angeles: U of California Press, 147-203. 474

Haiman, John (1971) Targets and paradigmatic borrowing in Romantsch. Language 47: 797-809. (1972) Phonological targets and unmarked structures. Language 48: 365-377. Kisseberth, Charles W. (1970) On the functional unity of phono- logical rules. Linguistic Inquiry 1: 291-306. Lightner, Theodore M. (1970) Why and how does vowel nasalization take place? Papers in Linguistics 2: 179-226.', Lovins, Julie B. (1971) Melodic conspiracies in L.3monv tonology. Papers from the Seventh Regional Meeting of the Chicago Linguistic Society. 469-478. Miller, Patricia Donegan (1972) Vowel neutralization and vowel reduction. Papers from the Eighth Regional Meeting of the Chicago Linguistic Society. 482-489. Nessly, Larry (1972) Conditioned natural processes in phonology. Paper given at the 1972 Summer LSA Meeting, Chapel Hill, North Carolina. Schachter, Paul (1969) Natural assimilation rules in Akan. IJAL 35: 342-355. Schourup, Lawrence C. (1972) Two aspects of vowel nasalization. Paper given at the 1972 Summer LSA Meeting, Chapel Hill, North Carolina.