The architecture of the english lexicon

1.2 Stress and prosodic constituents

The investigation of English stress in chapter three is built around the study of Kager (1989), who uses standard derivational accounts of English stress assignment to construct a system of groupings designed to categorize English words based on their stress patterns. Kager’s three principle groups differ in terms of the type of extrametricality effects seen at the right edge of the word: his first (and largest) group shows final syllable extrametricality (e.g., —rigin, appŽndix), his second group final consonant extrametricality (e.g., devŽlop, expŽct), and his third, "idiosyncratic" group, no extrametricality whatsoever (e.g., guit‡r, acquiŽsce). He also tries to broadly connect these groups with syntactic categories (although there are clearly many exceptions to this generalization), linking the first group to nouns and the second to verbs and adjectives. Kager also deals extensively with the issue of "retraction", which is based on the derivational assumption that stress in complex words is laid down iteratively as word formation progresses, and stresses can be assigned and removed during this process.

Kager’s decisions about identifying extrametricality and retraction stem from both the observed positions of main stress in the words as well as assumptions about foot-formation and derivation. His analysis conforms generally to the view that English displays the bimoraic trochaic foot (Hayes 1995), and the data in chapter three is presented in terms of such feet. Since the foot appears to play a role in both stress assignment, vowel alternation and many other phonological processes in English, it is relevant to briefly discuss the role of the foot and other members of the prosodic hierarchy in phonological theory. These prosodic constituents figure centrally in accounts of stress in both derivational theories and in OT, where they can act as arguments to various constraints.
1.2.1 The prosodic hierarchy and word-level stress

The prosodic hierarchy was originally developed to account for a wide range of phonological effects at the phrasal level (Selkirk 1980, Nespor & Vogel 1982). It was then extended to account for sub-word structure (Inkelas 1989, Cohn 1989), where it helped explain aspects of natural language such as stress which were difficult to express in terms of the purely segmental rules of Chomsky & Halle (1968). Prosodic constituents such as the prosodic word, foot, syllable and mora play critical roles in the various accounts of English stress discussed herein, both within derivational and Optimality theories.

In Chomsky & Halle’s system segments were the only available structural components and thus all features, including stress, were formally part of a segment’s feature inventory. Stress, however, was unusual in that it was the only non-binary feature, and furthermore it did not alter the identity of a phoneme when added to its feature matrix, in the way that other features like [voice] or [coronal] would. Stress also showed long-distance effects, the entire stress pattern of a word changing when certain suffixes were added. Stress rules were disjunctive and contained variables, which were not necessary for conventional feature-adjusting rules (Liberman & Prince 1977: 262).

The introduction of hierarchically organized tree structures and grids to account for stress acknowledged that "certain features of prosodic systems are not properties of segments (or syllables), but rather reflect a hierarchical rhythmic structuring that organizes the syllables, words, and syntactic phrases of a sentence." (Liberman & Prince 1977: 249) Following the crucial insight of Chomsky & Halle (1968) that heavy syllables tend to be stressed in English, syllable nodes were labeled "strong" if they "branched", i.e., if they contained a long vowel or were closed by a consonant (the classical metrical definition of "heavy" syllable weight). Thus, the same insight that had heretofore been expressed segmentally (by a disjunctive environment: VC or V[+tense] ) could be now formulated more elegantly in terms of branching in tree structure.

This approach was successful in eliminating the need for the complex stress rule system of Chomsky & Halle (1968). The n-ary feature [stress] was replaced by the binary feature [± stress], needed only to explain surface contrasts in unstressed vowels (e.g., ‡bsolute vs. Žxcellent), while the role of disjunctive ordering in the application of the stress rule was tied to morphological structure. Liberman & Prince’s (1977: 274-6) division of English words into (primarily) morphologically-based categories for "stress retraction" (following application of their universal English Stress Rule) is maintained by Kager (1989) (¤ 3.1.4). Membership in a particular retraction mode determined how foot-formation would take place, maintaining the link between stress assignment and extra-phonological information such as morphological structure and syntactic category contained within the analysis of Chomsky & Halle (1968).

The foot was introduced into the prosodic hierarchy as a formal prosodic constituent by Selkirk (1980), allowing for the successful jettisoning of the feature [stress], explaining all levels of stress via prominence relationships. Unstressed syllables were understood to be the weak members of feet, while full unstressed vowels were members of weak feet. Stressed syllables were the strong members of strong feet, and metrically heavy syllables could comprise feet by themselves. Primary stress was defined as a property of the prosodic word, which dominated the feet.

Selkirk’s ternary feet, used to explain the stress of words like Žxcellent, in which two syllables follow the main stress, were eliminated (at least for English²) by the proposal of extrametricality by Hayes (1981, 1982). This mechanism, which followed from the observation that ternary feet typically occurred at word-edges, allows for the elimination of peripheral phonological material from consideration by prosodic rules. This restricted the foot inventory to two types, binary and unbounded feet. The limited inventory of bounded foot types could be further distinguished according to specific parameters (Halle & Vergnaud 1978, 1987, Hayes 1995). These included various qualities which differed cross-linguistically across a limited range, such as the question of which syllable was dominant within the foot (i.e., an iambic or trochaic pattern) or whether the criteria for creating the feet from the underlying representation was quantity-sensitive. Quantity-sensitive systems, such as that of English, assign feet on the basis of syllable weight, heavy monosyllables forming feet of their own.³

Hayes (1985, 1987, 1995) further formalized the parameters for mapping the lexical entry onto a prosodic system, which included directionality, iterativity and persistence of parsing. His subsequent cross-linguistic studies (Hayes 1995) show that Hayes’ vision of the foot appears to be applicable over a wide range of languages. However, his use of extrametricality in deriving English stress has a large morphological component: many suffixes are marked as extrametrical, and extrametricality appears as a categorial feature of nouns but not verbs. Other rules, such as "Late Extrametricality", serve to mark the final syllables of unsuffixed adjectives, when "preceded by a branching foot" (Hayes 1987: 273) as extrametrical (e.g., d’fficult), effectively stating that adjectives with more than two syllables rather behave like nouns.

While Liberman & Prince’s (1977) concept of strong-weak branching constituents led to foot-based theories such as that of Hayes, their alternate representation of stress as a metrical grid led to the development of another approach, the rhythm-based framework used in Prince (1983), Selkirk (1984), and Nespor & Vogel (1986). These attempted to discard foot-like constituency and represent stress solely with a prominence grid. Rather than labeling members of a prosodic constituent as "weak" or "strong", Prince (1983) lays down grid marks which align to moras. These units of weight correspond to the presence of a heavy syllable (a long vowel or closed syllable) in other approaches, and so follow the insight previously captures by Chomsky & Halle’s stress assignment rule and Liberman & Prince’s "branching nodes." The latter’s insight that the grammar attempts to place stresses in a rhythmically alternating manner is represented by Prince’s rule of "perfect gridding", which iteratively assigns grid marks at a binary distance, modeling alternation.

The same principle can lead to destressing in clash, one instantiation of which was called "the rhythm rule" in Liberman & Prince (1977), which eliminates adjacent grid marks; i.e., when stresses fall too near to one another, one may be removed. Thus, syllables which we would expect to have a certain degree of stress fail to show it, due to the adjacency of another, stronger stress:

(1.2) thirtŽen Ü th“rteen mŽn p‡rent parŽntal
In Prince’s model, main stress was assigned by the End Rule, which added a superior grid level that encompassed the entire word, and he maintained Hayes’ principle of extrametricality to produce the proper environments for grid-mark assignment.

While this "pure grid" approach, which purposely eliminated constituency, was continued by Selkirk (1984), Hayes (1985) demonstrated that metrical constituency was crucial in accounting for phrasal rhythmic phenomena such as stress shift and rhythmic strengthening. Hammond (1984) proposed a model which placed the grid marks in a tree structure to represent constituency; this was further developed by Halle & Vergnaud (1987) who representationally conflated grids and feet into "bracketed grids". This formalism was later taken over into Hayes (1987), and his subsequent work. Thus, Halle & Vergnaud (1987) could represent both constituency and the hierarchical levels of grid marks indicating prominence, using a single representational schema.

To account for English stress, Halle and Vergnaud (1987: 277) assigned grid marks to syllables with a "branching rime", thus again capturing the insight that heavy syllables tend to be stressed. In the cases of some words which have unexpectedly stressed light syllables, such as medœlla, crebŽllum, they proposed lexical grid marks. Their "Alternator" rule performed the role of Prince’s "perfect gridding", modeling rhythmic alternation. Marks assigned at the first level of the grid acted as "heads" to the constituents which are thus defined. Halle & Vergnaud also relied heavily on extrametricality to eliminate final syllables from consideration, in the case of nouns and many affixed words. Further benefits of a grid-like analysis, such as destressing in clash and other rhythmic adjustments, were retained, but constrained by the use of constituency, thus incorporating the advantages of both the foot/tree and grid approaches.
1.2.2 Moras and minimality constraints

Hayes (1987) adopted the bracketed grids formalism of Hammond and Halle & Vergnaud, but also incorporated Prince’s use of moras, reinterpreting foot-assignment as the rhythmic grouping of bisyllabic (in quantity-insensitive languages) or bimoraic (in quantity-sensitive languages) units. In the latter type, heavy syllables were thus interpretable as feet in themselves, monosyllabic but bimoraic, whereas in Halle & Vergnaud’s formalism, they were assigned a grid mark simply on the basis of their branching rime. This placed the mora into the prosodic hierarchy, mediating between the syllable and the segments supplied by the lexicon and the morphology.

More evidence for constituency and a moraic analysis came out of the arguments for moraic conservation presented in Hayes (1989) and McCarthy and Prince (1986, 1988, 1990ab). Hayes demonstrated that moras, rather than timing tier slots,⁴ could representationally best account for the process known as compensatory lengthening. In many languages, moras whose melodic counterparts are deleted in phonological processes are unaffected by this deletion, but rather link to an adjacent melodic unit, which surfaces lengthened, conserving the moraic count within the word. Hayes (1995) gives further evidence for prosodic constituency, such as bounded stress migration, the retention of stress within the foot which occurs when stressed vowels are deleted during morphological processes.

McCarthy and Prince (1986, 1990ab) have shown that many reduplicative and templatic processes of affixation are only explicable through the use of prosodic constituents such as the syllable, foot and prosodic word. Furthermore, their observation of various lengthening processes in the context of such constituents led them to propose minimality constraints. They propose minimal word, foot and syllable sizes for languages like Arabic, Mokilese and Ponapean to account for a number of morphological processes cross-linguistically. Such constituents are then required by the grammar to contain a certain number of moras (usually two) or risk being ill-formed. The grammar consequently lengthens vowels or consonants inside those constituents when appropriate to achieve the necessary bimoraicism.

For example, as discussed by McCarthy & Prince (1990b: 21), in Modern Standard Arabic, vowels in monomoraic loan words from English are lengthened to achieve the bimoraic limit, e.g., baar ‘bar’, ïaaz ‘jazz’, gaaz ‘gas’. A Saudi Bedouin dialect geminates final consonants in the same situation, e.g., baêê ‘bus’, natt ‘nut’, rigg ‘rig’. In a different type of case, the reduplicative prefix in Mokilese is obligatorily a bimoraic syllable, regardless of the moraicism of the initial root syllable. Thus, light roots show a lengthened vowel in the reduplicative prefix; wad-wadek ‘read’, pok-poki ‘beat’, and caa-caak ‘bend’ might appear to simply repeat the form of their initial syllables, but paa-pa ‘weave’ and wii-wia ‘do’ show the lengthening required by the minimality constraint (McCarthy & Prince 1986: 20-1). Minimality is a good example of a constraint used within derivational theory, and its underlying concept is maintained in OT constraints such as FtBin ("Foot Binarity"), which will play a significant role in accounting for the lengthening processes described in chapter four.

Hayes (1987, 1995) incorporated the concept of quantitative minimality to his foot typology. The various types of binary feet are constrained to consist of two sub-foot units, either syllables or moras. For example, the moraic trochee, the foot-type found in English, is required to be bimoraic. Hayes’ typology of bounded feet⁵ is as follows (1995: 86):

(1.3) Syllabic Trochees: (s² s)

Moraic Trochees: (m² m) = (s² sù) or (s³)

Iambs: (m m²) = (sù s²) or (s³)
Hayes bases his principles of foot typology both on the psychology of the perception of iambic versus trochaic rhythms (Hayes 1985) and on the distribution of the various foot-types among the world’s languages (Hayes 1995). Hayes (1995) excludes from his inventory two types of feet allowed in previous theories, such as Hayes (1981) and Halle & Vergnaud (1987), the "even iamb" (s s²) and the "uneven trochee" (s³ sù), the latter of which is relevant for English. For cases where this foot appears to be necessitated by surface forms, such as in Latin domŽsticus, Hayes argues for parses like do(mŽst)ic rather than do(mŽstic), the penult being left unparsed. Other theories, such as Hayes (1981) and Halle & Vergnaud (1987), required all non-extrametrical syllables to be parsed into foot structure.
1.2.3 Stress assignment as foot-formation

Using metrical units like feet reduces the problem of stress assignment to that of foot-formation. Following the principles of foot-formation, based on a few parameters, underlying forms are parsed into constituents, primary and secondary stresses surfacing on those syllables that fall into the relevant head positions in the prosodic structure. Thus, the properties of stress are completely inherent in the prosodic constituents, and removed from the segmental level of analysis. Using such a prosodic hierarchy of constituents to account for stress within the word allows not only for a principled model for the apparent long-distance relationships seen in stress effects, but also provides environments, in the prosodic constituents themselves, for other phonological processes to occur, such as the vowel alternation effects discussed in chapter two.

The survey of foot-formation types in chapter three (¤ 3.2, ¤ 3.3) demonstrates that Kager’s (1989) ideas about the connections between syllable structure, syntactic category and foot-formation are simply rough generalizations which sometimes fail to hold over sizable groups of words. Evidence from sets of suffixed words (¤ 3.3.1) suggests that Kager’s extrametricality-based explanation of the contrast between his three major stress types is incorrect, and that instead, lexical structural marking on words of Kager’s third group (which, in his terms, fail to show any extrametricality effects) provide a better explanation of their behavior. This is similar to Halle & Vergnaud’s (1987) idea of lexical grid marks for words like medœlla (¤ 1.2.1), but here these are interpreted here as lexical moras, i.e., underlying geminate consonants. The presence of this additional mora in the lexical "input" form, which has no other surface effects, accounts for the unexpected stress patterns of these words, which display primary stress on apparently light final or penultimate syllables.

For Kager’s second group, identified with verbs and adjectives, the apparent lack of final syllable extrametricality will be accounted for herein by the proposal of a verbal/adjectival suffix / æ/, whose behavior is parallel to that of a class of Latinate suffixes including / al/, / or/, /-ous/. Identifying words of Kager’s second group as underlyingly suffixed with a non-surfacing final syllable allows for the apparent lack of extrametricality in these forms. It also accounts for the parallel stem shapes of pairs like prop—se Ü prop—sal, which can now both be understood as suffixed with metrically light monosyllables; these may be contrasted with the prosodic effects seen in other suffixed types (e.g., res’de Ü rŽsident, ins‡ne Ü ins‡nity). This is explained not by a blanket extrametricality restriction, but rather by proposing (¤ 4.2.2) that the suffixes / al/, / æ/ subcategorize for attachment to morphological stems differently than suffixes such as, e.g., / ity/, and this subcategorization has a concomitant effect on foot-formation. In chapter four, the relation between the foot and the stem will similarly be shown to have an effect on whether the stem vowel in words showing vowel alternation appears long or short in the output surface form.

1.3 Applying Optimality Theory

In chapter four, the observations and conclusions reached concerning the stress pattern of English in chapter three are interpreted using an Optimality Theoretic analysis. In this work, the OT approach is based upon Prince & Smolensky (1993), McCarthy & Prince (1993a, b, 1995) and subsequent work, but is constrained by a further theoretical goal: that the grammatical system should be limited, in the structures it can refer to, to members of the constituent hierarchies. This limitation is aimed at removing unstructured exception marking, special provisos, and other unconstrained extratheoretical mechanisms from a formal theory, forcing linguistic accounts to succeed or fail solely on the basis of whether the observed data can be accounted for using the formal structures and methods available to the theory. Subsets of data which cannot be accounted for by the theory should be understood as demanding improvement of the theory, rather than simply being labeled as exceptional.

While this study makes it abundantly clear that "exceptions" not only exist plentifully in the grammar of English, but in some cases embody sizable minority patterns, the theoretical interpretation of these minority groups needs to change. Rather than being seen as "exceptions" to absolute rules, these subsets can be understood as indicative of structural variations in the lexicon, and need to be accounted for using the same processes as the "regular" forms. This is possible using the novel approach found in Optimality Theory. In this work, OT will not simply be regarded as a "plug-in" replacement for the derivational cycle in phonology, but rather as the vehicle for providing an entirely new, unified understanding of how the grammar works. Many entrenched assumptions of previous analyses, such as the autonomous lexicon, the "basic" status of unaffixed forms, and the concept of "regular" words and "exceptions" are tied very closely to the mechanisms of derivational theory itself. OT provides a chance to re-evaluate these old assumptions and reinterpret the entire phonological system, from input to output form, using the rigorous principles of Optimality Theory.
1.3.1 Principles of Optimality Theory

There are very few theoretical mechanisms in Optimality theory, only constraints, candidates, and the ordering relationship captured by the hierarchical ranking of constraints. In terms of processes, a function termed Gen produces a set of all possible "output" candidates from the "input", which is usually taken to be a lexical entry, while a function Eval evaluates this set in respect to the constraint hierarchy. The ranking of a constraint hierarchy defines a particular grammar, and the candidate which minimally violates the constraint hierarchy constitutes the "output", a surface form (Prince & Smolensky 1993, McCarthy & Prince 1993a, b, 1995). As is conventional, Optimality analyses will be represented here as constraint tableaus, and will adhere to the notations used commonly in the OT literature.

McCarthy & Prince (1994: 3) offer a set of Optimality Theory principles, paraphrased here:
(1.4) Universality: Universal Grammar provides a set of universal constraints present in all grammars.

Violability: Constraints are violable but violation is minimal.

Ranking: Constraints are ranked on a language-particular basis; this ranking defines minimal violation.

Inclusiveness: The constraint hierarchy evaluates a set of candidate analyses that are admitted by general considerations of structural well-formedness.

Parallelism: Best-satisfaction of the constraint hierarchy is computed simultaneously over the whole hierarchy and the whole candidate set.
These principles sketch out the theoretical bases of OT, and can be addressed individually.

Universality refers to the set of constraints which may be used for evaluation of the candidates. For OT to be effective as a theory, this set must be limited and evidence for such constraints must be seen cross-linguistically. In this study, the constraints used all appear in some form in previous OT studies.⁶

Violability refers to the fact that constraints are not absolute, but can be violated by surface forms. This violation is mediated by ranking; the constraints are arranged in a hierarchy, and the candidate which least violates this constraint hierarchy is considered to be optimal.

Inclusiveness refers to the generation of all possible candidate forms by a function Gen. This aspect of the theory is especially relevant to the issue of the lexicon, as the candidate forms are based on an "input" form usually taken to correspond to the underlying lexical form of derivational approaches.

Parallelism refers to the simultaneous evaluation of candidates by the constraint hierarchy, in contrast to the serial derivation process of Lexical Phonology, in which words are built up step by step. While not all OT analyses maintain parallelism, not only can parallelism account for all effects in English usually ascribed to serial derivation, but theories based on serial derivation make incorrect predictions which parallel systems avoid. Simultaneous constraint evaluation provides a far more constrained and explanatory system than that of serial derivation, and parallelism will be maintained as necessary for a constrained Optimality Theory.

As a rigorous theoretical framework, Optimality Theory has the potential to produce more constrained and explanatory grammars than any competing linguistic approach. The complexity of natural language implies that a real-world grammar requires a very large (but finite) constraint hierarchy referring to all the relevant constituent structures seen in a given language. A single mis-ordering of constraints within a proposed constraint hierarchy will yield an incorrect grammar and make the wrong predictions for at least some words. Unlike the relatively risk-free addition of new, limited rules to a proposed grammar (in a derivational context) to account for difficult forms, adding new constraints to a system (or reordering constraints) for the same ends will also impact upon the evaluation of other words in the language. Only by presenting the precise constraint ranking which yields all the correct candidates will an OT analysis be a completely accurate representation of the grammar. Because of this precision, when such an ordering is arrived at, the arguments in favor of it being the correct characterization of the grammar are very strong. However, like any theoretical framework, OT is only useful when its structures and mechanisms can be clearly identified with and related to the data being evaluated.