Lexical Phonology and the History of English

This discussion of synchronic and diachronic phonology and their relationship is unsatisfactory for a number of reasons. The revisions McMahon proposes for lexical phonology are arbitrary and motivated more by the desire to reduce abstractness than to achieve explanatory adequacy. In her discussion of historical linguistics McMahon presents a gravely distorted picture of the view of early generative writers on this topic, particularly Kiparsky (1968) and King (1969), attributing to them claims which they expressly deny. She then presents as her own a view which differs little from these authors' actual views. She makes much of the apparent neglect of external evidence, but relies almost exclusively on internal evidence and ultimately obscures rather than clarifies the relation between the two. The result is a disappointment—I was expecting something much more interesting.

Lexical phonology (LP) offers principled solutions to a number of questions in phonology, morphology, and their interrelation that were problematic in Standard Generative Phonology (SGP). One of these is the restriction of cyclic rules to application only in derived environments, a restriction that explains the nonapplication of Trisyllabic Laxing (TSL) to nightingale without assuming the underlying segment /x/ of SPE, and also explains its nonapplication in stevedore, where SPE's solution using underlying /x/ is unavailable. McMahon apparently takes anti-abstractness to be a goal in itself (e.g., p. 7), in the expectation that this will facilitate the link-up with external evidence, and rejects the simplicity criterion; however, in practice her critique of the Halle and Mohanan's (1985, HM) version of LP makes little use of external evidence. Moreover, the pursuit of anti-abstractness for its own sake leads to gratuitous complications with no explanatory value.

For example, she criticizes the HM vowel system thus (p. 51): "If we were looking for a predominately surface-true vowel system, we would not find it here. The SPE account of Vowel Shift is retained ... Absolute neutralization is rife". That's odd: I thought we were looking for an explanatory theory. Actually, there is only one neutralization in the HM vowel system, the derivation of [jū] from both // and //. It is hardly "rife". And the Vowel Shift Rule (VSR) itself is not responsible for this or any other neutralization. It is interesting to recall that Kiparsky (1982a:130) specifically exempted Vowel Shift from the Alternation Condition for precisely this reason:

One of the effects of restricting phonology like this is to enter nonalternating forms in the lexicon in roughly their autonomous phonemic representation. That is, if a form appears in a constant shape, its underlying representation is that shape, except for what can be attributed to low-level, automatic phonetic processes. These can be defined as processes which do not cause neutralization of distinct representations. For example, the vowel shift of English, or the loss of final /g/ in sing, are low-level automatic phonetic processes, since the underlying form is in each case recoverable from the phonetic form.

Nevertheless, McMahon blames VSR for the abstractness of the HM vowel system, and proposes to cure this abstractness by moving it to the cyclic stratum 1, where it will be subject to the Derived Environment Condition (DEC) or Strict Cycle Condition (SCC), a descendant of the Alternation Condition. McMahon claims that HM "classify VSR as a noncyclic, Level 2 process, precisely in order to exempt it from DEC" (p. 90). I suspect their motivation was rather to find the best analysis. As Kiparsky (1982b:41) puts it:

The most important consequence of the Strict Cycle Condition is that there should be a relationship between the way a rule is ordered and the way it applies: cyclic ordering should correlate with the restriction to derived environments and postcyclic ordering should correlate with across-the-board application.

Indeed, subjecting VSR to the DEC condition does not have the explanatory force that this move has for TSL. It simply leads to enormous complications. Most directly, it requires splitting VSR into two rules. The original VSR remains, redubbed "Tense VSR", which applies in derived contexts such as Canadian. An inverted VSR must be added, dubbed "Lax VSR", which applies in derived contexts such as sanity. On this scheme, tense VSR cannot apply in sane, since this is not a derived context.

Her claim that both VSR are subject to the DEC encounters a serious problem when the underived form contains a neutralized vowel, such as harmony, with [ə] in the second syllable. She assumes (p. 102) an underlying // in RP and an underlying /α/ in GA, as they appear in harmonic, but this is a derived form, in violation of her own claim that the underlying vowel is that found in underived forms.

The splitting of Vowel Shift into two and placing it on stratum 1 entails numerous other complications, of which I will discuss only two. One involves Velar Softening, which in SGP and standard LP must be ordered before Vowel Shift in order to be formulated in terms of natural classes. Velar Softening must be ordered on stratum 2 because it cannot be subject to the DEC, since it applies in underived forms like reduce, oblige. To avoid this problem, McMahon simply denies that Velar Softening is a synchronic rule of English. This seems wrong—while it is not fully productive, it is fairly productive within a particular subset of the Latin-Greek derived vocabulary of English.

More interesting perhaps is McMahon's treatment of English strong verbs. Here she has missed the chance to identify a genuine deficiency in HM's analysis. HM propose several ablaut rules and a number of other rules for these forms, all of which they regard as phonological. This is a holdover from the SPE view of the interaction of syntax, morphology, and phonology, by which the phonology interprets the output of the syntax. In LP, by contrast, morphology takes place in the lexicon before words are inserted into syntactic representations. From this perspective, the ablaut rules should be regarded as morphological rules of stratum 1 (where other irregular morphology in English is located). This takes most of the punch out of McMahon's critique of HM for their use of minor phonological rules to account for this group of forms. HM's Lowering Ablaut accounts for a sizable group of English strong verbs such as (1).

(1)

1. a. sit, sat

2. b. eat, ate; lie, lay

3. c. choose, chose

We can regard Lowering Ablaut as a morphological rule marking vowels [+low] and marking the verb as past tense. This is straightforward in (1a); in (1b, c), the same morphological rule applies to mark past tense; (Tense) Vowel Shift must be allowed to apply in both underived and derived forms to get the right results. HM's Backing Ablaut accounts for verbs like those in (2).

(2)

1. a. spin, spun

2. b. find, found

3. c. break, broke

Our version of Backing Ablaut will make a vowel [+back] in conjunction with marking the verb past, and mark the vowel also [+round] if it starts off [-high]. This again produces the correct forms as long as Vowel Shift applies in underived find, break, as well as in derived found, broke. HM identify a third group of verbs exemplified in (3).

(3) write, wrote

Here the past tense is derived by applying both Lowering Ablaut and Backing Ablaut, in that order. From the morphological perspective, this amounts to a double marking, similar to a plural feets. This may be questionable. One type of analysis proposed by HM that is definitely disallowed in a morphological approach is their analysis of run, ran as having an underlying representation //, with Backing Ablaut applying in the present tense and Lowering Ablaut applying in the past tense. This is because the morphological approach views both ablaut rules as specific morphological markers of the past tense. Thus, unlike McMahon, who is forced to regard all the strong verbs as allomorphic,¹ we can allow some of the generalizations expressed by the ablaut rules to stand, but as morphological rather than phonological rules, while relegating some other verbs with greater irregularity (e.g., buy, bought) to the category of allomorphy.

McMahon claims that the SGP view of phonological change is "essentially static" (p. 9) citing Halle (1962), Postal (1968), and King (1969) as assuming

that a sound change, once implemented, is inserted as a phonological rule at the end of the native speaker's rule system; it moves gradually higher in the grammar as subsequent changes become the final rule [t]he assumption is that underlying representations will generally remain the same across time ... a sound change and the synchronic rule it is converted to will tend be identical (or at least very markedly similar), and the 'highest' rules in the grammar will usually correspond to the oldest changes.

Similarly, she states that "I am advocating that historical factors should be taken into account in the construction and evaluation of phonological models; but the mere equation of historical sound changes and synchronic phonological rules is not the way to go about it." Later (p. 11) she claims that "[t]his historical recapitulation contributes to further abstractness, and means that, in effect, related dialects must share common underlying forms. King (1969:102) explicitly states that external evidence, whether historical or from related dialects, may play no part in the evaluation of synchronic grammars; this is presented as a principled exclusion". She again cites King as advocating the necessity of historical recapitulation in synchronic grammars on pages 195–196.

It will be clear to anyone familiar with the SGP literature on sound change that this is a gross misrepresentation. It does not seem to bother McMahon in the slightest that the authors whom she cites as advocating this position in fact explicitly deny it. King (1969:102), for example, on the very page she cites as denying the relevance of external evidence, quite explicitly states a position exactly opposite to that which McMahon attributes to him on the questions of the equation of historical sound change with synchronic rules, historical recapitulation, and the commonality of underlying representations in related dialects:

Does historical evidence decide which of two synchronic grammars is higher valued? The answer is a flat No. Given two grammars G₁ and G₂ that correctly account for the same data, and given that G₁ is simpler than G₂ but that G₂ more nearly recapitulates the historical development, then the simpler grammar G₁ is higher valued than G₂. Given two grammars G₁ and G₂ of equal simplicity, and given that G₂ better reflects the historical development of the language, nevertheless both grammars are equally valued in the evaluation measure ... Why evidence of these types is not directly relevant to the evaluation of synchronic grammars should be clear. A grammar is an account of a speaker's intrinsic knowledge of his language, his competence—not his father's competence, not any of his ancestors' competences, not the competence of his neighbor whose dialect is slightly different ... if historical information is allowed to enter into the evaluation of synchronic grammars, the whole question of the relationship between a synchronic grammar and its history ceases to be of any interest. As it stands, it is an interesting, empirical question whether a particular grammar recapitulates history ...

Similarly, Halle (1962:65) states that "[t]he addition of rules may—though it need not invariably—result in a grammar that is not optimal (the simplest) for the set of utterances that it generates" and (1962:65) "Since every child constructs his own optimal grammar by induction from the utterances to which he has been exposed, it is not necessary that the child and his parents have identical grammars, for ... a given set of utterances can be generated by more than one grammar." And Kiparsky (1982a:130) puts it this way:

It is a very natural, though theoretically unjustified, desire to have synchronic descriptions reflect diachrony to the greatest possible extent. The greater the similarity between synchronic and historical grammar, the less work either of them involves for the linguist. It would be ideal if we could simply provide the arrowheads of historical grammars with shafts to get synchronic descriptions, and perform the converse operation on The Sound Pattern of English to get a history of English phonology. But unfortunately we cannot assume that synchronic grammars necessarily have a form which takes the hard work out of internal reconstruction. Children learning their native language do not have the interests of linguists at heart.

To be sure, McMahon allows that "there are occasional cases of rule loss or rule inversion" or "that some limited provision is made in SGP for the restructuring of underlying representations" (p. 9), but on the whole she claims that, in SGP, "rule addition ... is the main mechanism for introducing the results of change into the synchronic grammar ... SGP is essentially a static model" (p. 9) and "[r]estructuring of the underlying representations during acquisition by later generations of speakers is theoretically permitted, but infrequently invoked" (p. 196). In McMahon's caricature of the SGP view of sound change, the synchronic grammar of any language consists of all the sound changes that have occurred in that language, in the order in which they were added to the grammar. As she puts it, "the historical phonology of a language will be almost directly mirrored in the order of its synchronic phonological rules" (p. 196). In this view, the grammar becomes ever more complex with the passage of time, with static underlying representations and ever more and more rules. As we have seen, the actual practitioners of SGP deny this view. King (1969) gives the clearest, most detailed, and most explicit explanation of the SGP view of sound change. He devotes an entire chapter to grammar simplification. He emphasizes that language change is grammar change,² and that, while rule addition is the major way in which adult grammars change, children's grammars will frequently exhibit rule loss, rule reordering, and restructuring, that is more than "occasional" or "limited". While grammatical theories have changed substantially since 1969, these ideas still seem to lie at the foundation of any generative description of language change.

On the positive side, McMahon presents two case studies in terms of her approach. Chapter 4 discusses the Scottish Vowel Length Rule (SVLR), its historical evolution, and interaction with the Great Vowel Shift and other rules. Part of the discussion revolves around which of vowel length or tenseness (or both) are distinctive in underlying representations in English. Oversimplifying considerably, RP and GA are claimed to have an underlying distinction in length and and perhaps tenseness also, while Scots and SSE are claimed to have only a tenseness distinction, thus supporting her contention that these two dialect areas differ in underlying representations (though, as I have already observed, this possibility is allowed in earlier discussions of LP and SGP). Much of the discussion is theory internal, with the shift to underlying length (e.g., HM) from underlying tenseness (SPE) influenced partly by the adoption of autosegmental and metrical rather than linear representations. In addition, she discusses the life cycle of the SVLR, from its innovation as a postlexical rule, through its evolution into a lexical rule, with some of its effects later incorporated into underlying representations. The application of this idea to SVLR is novel, but the possibility of such a life cycle was already proposed in Kiparsky (1982b) and so is not new.

Chapter 6 is devoted to English r in nonrhotic dialects. McMahon rejects McCarthy's (1991) solution that involves both insertion and deletion of r, proposing that insertion alone is sufficient. This causes difficulties with alternations like severe ~ severity, which, according to her restrictions, has an underlying stem /səviə/. In severity, she claims that r-Insertion must precede Trisyllabic Laxing and thus be ordered on stratum 1. This in turn implies that r-Insertion will apply in stratum 1 derivations such as algebraic, which is incorrect. McMahon makes the extremely important observation that most words of this type, such as operatic, have an allomorph -tic of the suffix (it could also be an allomorph operat- of the stem). Nevertheless, words like algebraic are still a problem for her. In addition, algebra will have to have schwa as the underlying final vowel in her system, raising the problem already mentioned in the case of harmony above; that is, the problem of determining which full vowel should emerge from an underlying reduced vowel when this gets stressed (and tensed), as in algebraic. McMahon cites some apparent cases of r-Insertion before level 1 suffixes, such as Sha[r]ify, without giving a source; I find these rather strange, but I am not a native speaker of this dialect. R-Insertion must apply postlexically in any case to account for phrasal cases. Its stratum 1 application seems to be more a consequence of her ordering of Vowel Shift on that level than of any empirical considerations. It seems inconsistent that she is so anxious to have both the historical and an inverted form of Vowel Shift but vigorously rejects a similar analysis for r-insertion and deletion as proposed by McCarthy.

Chapter 5 is devoted to an unsuccessful attempt to argue against underspecification. McMahon begins by claiming that underspecification permits related dialects to share underlying representations where she would analyze them as having distinct underlying representations, then claims that underspecification makes it easier to relate dialects by assigning them the same underlying representations. However, these two issues are logically independent of each other—either the assumption of related dialects sharing underlying representations or the opposite assumption is consistent with either underspecification or full specification. Summing up, the strongest thing she can say is that "there may be machinery in linguistic theory that leaves doors open for us which are better off closed. Radical underspecification may be one such piece of apparatus" (p. 227, emphasis added). Hardly a devastating critique! She cites McCarthy and Taub's (1992) claim that coronal underspecification in English encounters difficulties in that several processes need to refer to coronal specifications. However, McCarthy and Taub's claim is simply incorrect: the mere existence of rules referring to default values does not undermine underspecification theory. Default values need to be filled in at some point in the derivation; Archangeli (1984) has proposed the Redundancy Rule Ordering Constraint (RROC) to ensure that default values are filled in before any phonological rule refers to them. Under this assumption there is simply no problem. This restriction is needed independently to ensure that binary features do not have ternary power by allowing the unspecified value to function as distinct from plus and minus. Underspecification theory also needs to ensure that only one value (either plus or minus) of any feature can appear in underlying representations in any given environment (Kiparsky 1982b:54).

McMahon misinterprets Archangeli's (1988:193) discussion of the learnability of a radically underspecified system. Depending on the system of a language, either + or - can be the specified value of a given feature, with the other supplied by default. Archangeli states that "[i]f such options are freely available, the learnability of a system becomes quite a challenge" (quoted by McMahon on p. 218), but this hardly leads to a "theoretically unlimited number of underlying systems for any set of surface forms" (p. 218), as McMahon imagines. First, since each feature is limited to two possible specified values, there can be at most 2ⁿ possible analyses of a system defined by n binary features. More importantly, Archangeli continues on the same page in a passsage not quoted by McMahon:

Thus, some principle must be included in the theory in order to reduce the burden on simply learning the inventory. The proposal in Archangeli (1984) is that certain rules/specifications are preferred by Universal Grammar. During acquisition the first approximation will be in accord with these universal preferences and other options will be selected only if language-particular evidence motivates such variation.

This refutes McMahon's belief in "the doubts raised by practitioners of underspecification over the learnability of any radically underspecified representation" (p. 227, again citing Archangeli 1988:193).

McMahon regards several of Borowsky's (1986) uses of underspecification to be diacritics that allow free rides. One such instance is an analysis of the Sanskrit ruki rule, which is actually due to Kiparsky (198b 2:75–82). This rule converts /s/ to a retroflex [ṣ] when it follows one of the segments /r, u, k, i/. The rule applies in derived environments. A retroflex [ṣ] is the normal situation after the same segments in underived contexts as well; however there are exceptions where unretroflexed [s] appears after those segments in underived contexts and a few cases where a retroflex [ṣ] appears in a non-ruki environment in underived forms. Kiparsky points out that, in any given context, only one specified value of the retroflex feature need be included in underlying representations and that there is no ternary power. The specification [+R] appears only in underived nonruki contexts, such as /ṣaṣ/ 'six' and [-R] occurs only in underived ruki contexts that are exceptions to ruki, such as /barsa/ 'tip'. Otherwise, /s/ is unspecified for [R], receiving the specification [+R] in ruki contexts and [-R] otherwise by default. The major advantage of this analysis is that it provides a solution to the duplication problem: the situation where a Morpheme Structure Rule (or Morpheme Structure Condition) duplicates a phonological rule. Given underspecification, the same rule can function as a Morpheme Structure Rule, filling in a specification in underived contexts, and as a phonological rule in derived contexts. McMahon disregards this result of underspecification, apparently preferring to let such generalizations to go unexpressed.

There are relatively few typographical errors, as expected from Cambridge, but a few need to be noted. Idiosyncrasy is consistently misspelled with c instead of s in the last syllable. Diagram 3.11 on page 97 illustrating the operation of "Lax vowel shift" has the delinking bars on the first, rather than the second, of the two timing slots after the first arrow. Most seriously, throughout Chapter 3 the phonetic symbols have undergone a font change, so that for example [α] appears throughout as [¡], and numerous other phonetic symbols have similar odd substitutions. The results are decipherable to anyone familiar with these discussions, but will appear as absolute gibberish to newcomers. On page 158, diagram 4.15 illustrating stage two of the Great Vowel Shift has few as an illustration of a change [u] → [eu]; it is not clear what is intended, since few has [] in Middle English, which shifts to [yu] in Early Modern English, a change in syllabicity within the diphthong rather than a shift in vowel height as with the other changes illustrated in the diagram.

In sum, this book claims to make contributions to Lexical Phonology and to Historical Linguistics. The new proposal for Lexical Phonology is novel, but unconvincing: it comes from the unmotivated and arbitrary claim that even nonneutralizing rules like Vowel Shift should be subject to the Strict Cycle Condition, in the absence of any evidence that such rules apply cyclically. The proposals for historical linguistics are valid, but not novel, since Kiparsky (1968) and King (1969) long ago accepted the idea that related dialects can differ in their underlying representations and that such representations are restructured as soon as synchronic evidence for the rules that resulted in the innovations becomes unavailable to the learner. No one should rely on this book as a source of information on earlier scholarship in the area; it must be read in conjunction with such earlier sources in order to get a true picture of the history of these ideas.