Review of Giegerich, Heinz J. 1999. Lexical Strata in English: Morphological Causes, Phonological Effects. Cambridge Studies in Linguistics 89. Cambridge: Cambridge University Press. $69.95.
Reviewed by Graham Horwood, Rutgers University
The author's short introduction presents the work as the reanimation of a dead framework, arguing that in fact, despite criticisms levied against Lexical Phonology (LP henceforth) in the late eighties, a sizeable body of the framework's achievements are salvageable under a subtle realignment of the definitional basis of 'strata'. Giegerich (HG henceforth) argues that the critical failing-point of LP was the affix-driven stratification hypothesis. This model presumes that the morphology of a given stratum is defined by the collective behaviors of diacritically-marked affixes. HG observes that the existence of 'dual-stata' affixes, those which display characteristics consistent with more than one stratum, undermines any coherent notion of what a 'stratum' is in the first place. HG's solution to the dilemma comes in the theory of 'base-driven' stratification, the driving theoretical force behind the remainder of the work. If strata are defined with respect to the characteristics of affixational bases (roots, stems, words), polystratal affixes cease to undermine the theory, and a number of problematic cases for traditional LP receive a relatively straightforward account.
The second chapter elaborates on the short-comings of the affix-driven stratification theory. The author first outlines in detail the various diagnostics proposed in past literatures for determining the stratal classification of various morphemes. HG goes on to catalog the fourteen derivational suffixes of English which defy simple categorization by these diagnostics, demonstrating behaviors endemic of both stratum-1 and stratum-2. The commonly cited suffix -able/-ible, for instance, occurs in doublets such as tolerable/toleratable; the semantic non-compositionality of the former ("moderately good") is indicative of stratum one behavior, while the semantic transparency of the latter ("able to tbe tolerated") is stereotypical of stratum-2. It is necessary to mention that HG does put a considerable amount of explanatory burden on such forms as 'gormless', an example which *might* prove that an otherwise well behaved stratum-2 suffix like -less can demonstrate the canonically stratum-1 property of affixing to a bound root. It might also demonstrate that lexicalized, morphologically simple (and largely uninformative) forms may bear a remarkable phonological resemblance to morphologically complex ones. HG concludes that the groups of phonological, morphological, and semantic characteristics which define a stratum are not in fact discrete, but may overlap for some morphemes.
HG demonstrates a number of other failings of affix-driven stratification in Chapter 3, going on to argue that, collectively, the weaknesses of the theory serve as an indicator of the changes which LP must undergo in order to remain a viable framework. First and foremost, HG observes that many of the predictions made in the affix-based model by the Affix Ordering Generalization (Selkirk 1982; AOG henceforth)--no stratum-2 affix can occur inside a stratum one boundary, but the opposite may occur freely-- are in fact ruled out by idiosyncratic facts of individual affixes as argued by Fabb (1988). Words such as *sensiblize, for instance, are to be expected under the AOG. Furthermore, many of the ordering facts originally accounted for by the AOG are equally well explained by the observation that affixes diacritically marked [+Latinate] never attach to forms marked [-Latinate]. The fact that this [+/-Latinate] constraint holds equally well for both simplex and complex words (for example, *shortity and *homelessity) suggests that in this domain at least a stratal split must be made according to properties of *bases* rather than *affixes*. Taking this observation to a higher logical plane, HG argues that in fact stratal categorizations are reducible to the lexically listed diacritical markings of the base to which a morpheme attaches.
On stratum-1, each morpheme (simple root or affix) is coded to directly select for the affix(es) which may attach to it. Thus HG accounts for: a) the fact that a majority of traditionally stratum-1 affixed forms are relatively unproductive and semantically noncompositional; b) the ill-formedness of *sensiblize; and c) the robustness of the [+/-Latinate] constraint. Forms such as [gorm-] select for [-less]; [short] selects for [-ness], not [-ity]; [sense] selects for [-ible], but [-ible] does not select for [-ize]. The empirical coverage of the model is quite broad. As it result from lexical stipulation rather than grammatical principle, however, the possibility of any formulization of generalizations regarding the ordering of affixes, including the [+/-Latinate] constraint, is lost. HG is undaunted by this fact, pointing to such examples as 'withdrawal' and 'winterise' to show that in fact that the [+/-Latinate] constraint is not so inviolable as previously suggested. This reviewer is independently aware of further evidence for such an analysis; in the Australian Aboriginal Language, Kayardild (Evans 1996), suppletive allomorphy in multiply-suffixed forms is conditioned by the form of any immediately adjacent morpheme (be it the root or another affix), rather than just the declension of the root. I mention this fact to lend some support to the author's somewhat stipulative analysis, as well as to draw attention to what I take to be a deficit of the book. While obviously the primary matter of a work entitled _Lexical Strata in English_ must be the English language, HG's apparent refusal to weigh his assumptions and conclusions against the facts of other languages (except of course German, hardly a language in dramatic typological contrast with English) leaves the reader to wonder at the universality of his claims.
The crucial distinction between stratum-1 and stratum-2 affixes proves to be a more troublesome aspect of HG's model. Traditionally, one of the distinguishing properties of stratum-1 and -2 affixes is that the former may attach to bound roots unspecified for lexical category (such as 'gorm-'), while the latter may not. This distinction is captured in HG's system with the assumption that all stratum-1 morphemes (even morphologically complex "roots" derived through several stratum-1 affixations) are unspecified for lexical category, while all stratum-2 forms are fully specified as words; roots acquire category information in the process of becoming words, i.e., undergoing a root-to-word conversion rule. With this assumption comes a readily obvious dilemma: how are robustly attested selectional restrictions of stratum-1 suffixes to be captured (-able attaches to verbs, but not nouns, for instance)? HG answers this question with the assumption that roots--while free of categorial information proper--are diacritically marked for the category they will become after root-to-word conversion. This crucial distinction between categorially-unspecified-root-which-will-become-a-noun- after-root-to-word-conversion and noun seems somewhat suspicious.
HG's root-to-word conversion rule is shown in Chapter 4 to serve double duty in the proposed framework, not just acting as a formal bridge between root and word strata, but also subsuming the utility of the 'identity rules' of Kiparsky (1982). For underived words, root-to-word conversion--disjunctively ordered with structure changing rules under Kiparsky's (ibid.) Elsewhere Condition--constitutes a more lexically specialized (every base is listed with its appropriate conversion rule) stratum one process capable of blocking more general structure changing rules. As a consequence, so-called Strict Cylicity Effects (SCE) are divorced from cyclicity proper, and hinge instead on the stratal location of a form at a given point in the derivation: SCE holds for all cycles in all non-final strata. This is a welcome result for HG, who goes on to show that, resultingly, any need to arbitrarily designate a "pre-morphology" cycle for free roots (to the exclusion of bound roots) is obviated, and similarly the need for 'free-ride' derivations, wherein never-surfacing underlying forms are posited solely for the reduction of rule complexity, are no longer necessary in a number of oft-discussed problems of LP, including English Vowel Shift and mn-Simplification. The investigations of the remaining chapters of the book are motivated primarily by the author's desire to rid LP of such analyses.
In Chapter 5, the author attempts to argue on psycholinguistic grounds that grammatically costly derivations are preferable to more grammatically economical analyses. Traditionally, analyses of derivationally related pairs such as 'atom - atomic' have non-controversially relied on some rule of Vowel Reduction to operate on the morphologically simplex form, reducing its second-syllable vowel to schwa. HG argues that analyses such as these are ill founded as they rely on the (apparently ridiculous) assumption that language learners must hear the morphologically complex form and associate it with the simplex form before its underlying full vowel can be posited. The incidentals of the author's admittedly *more* complex alternative analysis are not key to discussion here; some objection to the author's final conclusions, however, must be raised. It is rightly observed that written forms can influence pronunciation in adult production grammars. The author's insistence, however, that "Only the literate speaker can establish a regular derivational path from 'atom' to 'atomic'" (166) is a questionable conclusion about the linguistic competence of English speakers generally. Based on this statement, the author must conclude that members of the English speaking world's very large percentage of illiterate speakers make no derivational connection between such forms, memorize their surface forms by rote, and in fact possess phonological grammars completely free of the complex system of phonological rules alluded to above. Furthermore, since the rote listing of stratum-1 forms is taken to be one of the commendable points of the author's system in Chapter 3, it remains to be seen why the language acquisition system of *literate* speakers would burden itself with such a complex grammatical architecture when memorization is an equally tenable option.
Chapters 6 and 7 provide in-depth analysis of [r]-sandhi phenomena in dialects of British English--for example alternations such as 'hea - hea[r]ing' in RP (Received Pronunciation). Paralleling it with [j]- and [w]-sandhi in examples like 'sa[y] it' and 'sho[w] it', HG takes [r]-sandhi to be an effect of liaison, resulting as an underlyingly unspecified nuclear melody is filled by postlexical default rules. Various arguments, diachronic, synchronic, and dialectal, are given for this treatment of the alternating segment as unspecified for [+/-consonantal].
HG's treatment of liaison leads naturally into discussion of syllabification in Chapter 8. The principle deviation HG makes from standard LP here lies in his argument that there is no need for structure-changing resyllabification rules in the proposed model. Challenges to this argument come in comparison of syllabic-sonorant alternations such as 'hinder - hindrance', as well as forms showing liaison. HG's approach to such examples is to formulate his stratum-1 syllabification rules in such a way as to leave word-final consonants unsyllabified until stratum-2, where a special 'Rhyme Condition' rule syllabifies final sonorants as rhymes; stratum-1 affixed forms, such as 'hindrance', will avoid the rule, while stratum-2 affixations will show its effects, as found with the stratum-2 gerundive suffix in 'hindering'. HG's treats of exceptions to the rule--for example 'kindling' (i.e., wood used to start a fire)--as being reanalyzed from stratum-2 affixation to stratum-1. This appears to be one of the less restrictive aspects of his framework. Since affixes are no longer tied to particular strata by dint of their collective phonological behaviors, it obtains that *any* given counter-example to one of HG's assertions about the nature of stratum-2 morphology can simply be dismissed to stratum-1 through "reanalysis".
Overall, the work is worthy of a tentatively positive review. On the negative side, the author does not at any point attempt to evaluate his LP against the findings of more current theories of morpho-phonology. This somewhat cloistered theoretical standpoint does detract considerably from the work's utility. However, for the student of morpho-phonology not so fortunate as to be intimately involved in the flurry of work in LP which emerged in the '80s and early '90s, the book serves as an excellent introduction of the intricacies of that theory. HG's overview of and comparison with the earlier theory is sufficiently general to facilitate understanding of many of the foundational assumptions and achievements of LP, and his own innovations regarding it do possess some insight beyond the limitations of the chosen framework. In the earlier chapters particularly, HG's work is more a consideration of the underlying theory of morphological organization behind LP than one of the precise formulation of the phonological rules of English. As such, it is valuable to the morpho-phonologist working under the assumptions of any given framework.
Selkirk, Elisabeth. 1982. _The syntax of words_ (Linguistic Inquiry Monographs; 7). Cambridge, MA: MIT Press.
Fabb, Nigel. 1988. "English suffixation is constrained only by selectional restrictions", Natural Language and Linguistic Theory 6: 527-539.
Evans, Nick. 1996. _A grammar of Kayardild : with historical-comparative notes on Tangkic_. New York: M. de Gruyter.
Kiparsky, Paul. 1982. "Lexical phonology and morphology." Pages 3-91 in I.S. Yang Linguistics in the morning calm, volume 2. Seoul: Hanshin.
Graham Horwood is a Ph.D. student at Rutgers University. His current work explores processual morphology and the implications of lexical-network based morphological architectures for Optimality Theory.