in recent phonological work has been the elimination of structure-changing processes from the grammar." With structure-changing process gone, all that conAttention on constraint-based grammar formalisms straints do is specify what it means for surface forms such as Head-driven Phrase Structure Grammar (HPS6) to be well-formed 2. A consequence ofthis paradigm has focussed on syntax and semantics to the exclushift in phonology is that he main barrier to the incorsion of phonology. This paper investigates the inporation of phonology into constraint-based grammar corporation of a non-procedural theory of phonology frameworks has now disappeared. into rmSG, based on the 'one-level' model of Bird & Ellison (1992). The standard rule-representation In trying to assess the consequences ofthis prodistinction is replaced by the description-object disgramme for phonology itself, it would be instructive tinction which is more germane in the context of to identify four major ideas or principles that have constraint-based grammar. Prosodic domains, which characterised work in the Montague tradition, out limit the applicability of phonological constraints, are of which constraint-based grammar formalisms have expressed in a prosodic type hierarchy modelled on grown. The first is THE PRINCIPLE OF COMPOSITIONlfPSG'S lexical type hierarchy. Interactions between ALITY, which requires that the form and interpretation phonology and morphology and between phonology of an expression be a function of the form and interand syntax are discussed and exemplified I . pretation of its parts. This principle motivated Bach's well-known RULE-TO-RULE HYPOTItESIS, requiring that the linguistic modules (such as syntax, semantics and phonology) be organised in parallel rather than in series, where the output of one module is the input of the next. The second principle is Partee's The exclusion of phonology from Head-driven Phrase (1979, 276) WELL-FORMEDNESS CONSTRAINT, which Structure Grammar (Pollard & Sag, 1987) and restates that well-formed expressions are buih up out lated grammar frameworks belonging to the Monof well-formed parts. An immediate consequence tague tradition (Dowty et al., 1981; Shieber, 1986) of this is that there can be no phonological 'repair is no doubt due to the perception that much curstrategies' which operate on ill-formed structures. rent theorising in phonology is excessively procedural in nature. While there is an element of truth in A third principle goes by the linked terms INTENSIONthis, it is also true that the predilection of constraintALITY and PARTIALITY. The idea here is that there is based grammarians for a string-based phonology has a distinction to be drawn between linguistic descrippredisposed them towards segmental phonology, in tions and the objects they describe (such as pieces which procedural thinking is at its peak. Other variof speech). Descriptions can only ever be partial, eties of phonology--most notably non-linear phonolin the sense that they pick out---or DENOTE----classes ogy (Goldsmith, 1976, et seq.)--have now largely of objects rather than individual objects. For examabandoned complex derivational accounts of phonople, in a given language, \[+voice\] denotes a class of logical well-formedness in favour of systems of genspeech sounds, \[+nasal\] denotes another class, and eralised constraints. A defining property of con\[+voice,+nasal\] denotes the intersection of these two straints is that they are not applied in any particular 2Historical note: There is a close parallel between this theoorder. Rice (1989, 331) writes that "a general goal retical position and that adopted by Th~o Venneman, Joan Hooper and Grover Hudson in the 70's in the theoretical framework IThis research is funded by the U.K. Science and Engineerknown as Natural Generative Phonology (Hooper, 1976). More ing Research Council, under grant GR/G-22084 Computational recently, Bach (1983) and Wheeler (1981) have independently Phonology: A Constraim-Based Approach. I am grateful to adopted asimilar stance regarding the incorporation fphonolMark Ellison, Ewan Klein and Bob Ladd for discussions relating ogy into categorial grammar. to this work. ACTES DE COLING-92, NANTES, 23-28 AO'~r 1992 7 4 PROC. OF COLING-92, NANTES, AUG. 23-28, 1992
classes. Refining a description results in a smaller class of denoted objects 3. The final principle is TIlE LF2OCALIST tP/POTIIESIS as expounded in its strong form by Bresnan and others, which has lead to the lexicalisation of linguistic information and the corresponding simplification of phrase structure roles. From these themes come such requirements hat there be no extrinsic rule ordering, no 'feature-changing', no intermediate levels of representation, and so on; all of them principles that many phonologists have independently come to embrace. The ultimate implications of this perspective for phonology is the removal of the role-representation distinction in favour of the description-object distinction 4.Grammar formalisms like IIPSG also lack the rule-representation distinction. For exanlple, (1 a) is described by Pollard & Sag (1987, 149) as a role. However, they also provide an equivalent statement of the rule in the more conventional 'rewrite' notation (lb). compass nonlinear phonology (Kay, 1987; Koroai, (1) a. \[it~t>1n RIsYNlUx:ItJaX mrs \[COMe-DTRS (\[l) b. \[SUBCAT 0\] ~ IIILFX 'l, (2
In this paper a constraint-based phonological framework is presented. It is shown how various interactions between phonology and the other modules of grammar can be studied within the context of \[IPSG. Section 2 presents the finite-state model of phonology advanced by Bird & Ellison (1992). This is followed by a section on a phonology-morphology interaction (nasalisation in TerSna) and a section on a phonology-syntax interaction (complex NP shift). The paper concludes with a discussion of the shortcomings and future prospects of this work. 2 Finite-State Phonology Over the last decade much has been written on the application of finite-state transducers (FSTs) to phonol-
~This modelqheoretic viewpoint is explored in detail by Bird (1990). A first-order language, similar to Johnson's (1988) attribute-value language, is presented in which both phonological Stnletures and feature structures can be expressed. Bird & Klein (1990, 53), Pierrehumbert (1990, 392) and Ilird & Ladd (1991, 206-7) discuss ome ramifications of this view of phonology.
4This new view of phonology has been investigated by Bird (t 990), Coleman (1991). Scobbie (1991) and by the contributors to the collection (\]Bird, 1991a). A¢nES DI! COLING-92, NANrEs, 23-28 Ao~r 1992 7 5 ogy, centering on the TWO-LEVEL MODEl. of Koskenniemi (1983). Antworth (1990) gives a comprehensive introduction to the field. The two-level formalism is an attractive computational model for 1960's generative phonology. However, as has "already been noted, phonologists have since moved away from complex string rewriting systems to a range of socalled nmflinear models of phonology. The central innovation of this more recent work is the idea that phonological representations are not strings of characters but collections of strings, synchmnised like an orchestral score. There have been some notable recent attempts to rescue the FST model from its linearity in order to en1991). llowever, if a fundamental shift in perspective in phonology has indeed occurred, then these refinements to the FST model do not go far enough. We require a further restriction that a transducer can only add information: the set of symbols accepted on a particular cell of the 'surface' tape must be a subset of the set accepted on the corresponding cell of the 'lexical' tape. An I;ST so constrained is actually nothing more than a finite-state automaton (FSA).
Consequently, FSAs would now seem to be more appropriate than FSTs for doing phonology, so long as it is possible to come up with some way of encoding nonlinear phonology using FSAs. In fact, this has been done by Bird & Ellison (1992), who have shown how autosegmental TIF.RS, ASSOCIATIONS and RULES can be encompassed by FSAs. The central idea is that each tier is a partial description of a string, and tiers are put together using the intersection operation (vI) defined on FSAs. Apart from being truer to current phonological theorizing, this one-level model has a second important advantage over the two-level model. Since the set of FSAs fomls a boolemt lattice under intersection, union and complement, we can safely conjoin ('unify'), disjoin and negate phonological descriptions. Such a framework is obviously compatible with constraint-based grammar fomudisms. As it happens, tim currant usage of phonology in I1PSG is just a simple form of finite-state phonology of the kind elatx~rated here.
As our lirst example, consider the phenomenon of homorganic nasal assimilation, wherehy nasals agree in place-of-articulation with the following consonant. Thus, the sequences \[mb\] and \[ndl are allowed while lind\] attd \[nb\] are ruled out. 1 ,et N = {m,n}, S = {b,d}, L = {re,b} and A = {n,d}. The required constraint can be expressed as the following regular expression, Prtoc. ov COLING-92, NANTt!S, AUG. 23-28, 1992 phon-domain hhr!eYC'~hhr~e~ilght~ i yl ~f,a°°t ~e~/i~t~.. Figure 1 : A Prosodic Type Hierarchy where ' ' stands for any character and '*' is the Kteene star. .*(NS cq LL IA AA).* This states that it is not possible to lind anywhere a nasal-stop cluster (NS) which is not made up of two labials (LL) or two alveolars (AA). We can simplify the above expression as follows: (2) .*(mA)." VI .*(nl,)fi The lirst conjunct of (2) corresponds to the following NFA (all states are both start states and end states): rh m rnnA This automaton will accept any string, so long as every \[m\] is followed by another \[ml or by a ramalveolar (A). The wlmle of expression (2) cormsponds to the following NFA (where again, all states are both start and end states): ~q @ m • _ < _ _ _ s~x// m ~'/ ()m e ~ t l \[I (5 n
Before proceeding further with this example, it is necessary to define a PROSODIC TYPI- IIIERARCIIY. AcrEs I)E COLING-92, NANrEs, 23-28 Aot)r 1992 7 6 PRO(:. OF COLING-92, NaN'it/s, AUG. 23 28, 1992 This is a subsumption etwork akin to the lexical hierarchy of ttPSG (Pollard & Sag, 1987, 191ff). Figure 1 displays a simple prosodic hierarchy. Each of the prosodic types denotes a set of phonetic objects (± denotes the empty set). Since our descriptions are expressed using regular expressions, it is useinhabiting ful to imagine regular expressions (or automata) as he nodes of the prosodic hierarchy. In English, nasal assimilation dries not occur across phrase bxmndafies and so the automaton given above must only constrain those stretches of segments that are phrases. This localisation of the phonological constraint could be expressed as follows: (3) (~ m ,@m t 2:?? NnL n n phrase C n
The tnp element in Figure l, namely phon-domain, denotes the space of utterances 5. Phonological constraints like (3) serve to restrict his space, leaving just those utterances that are acceptable in the language. The phonological specitication in any sign (including a lexical sign) is a member of this hierarchy.
In some langttages we may require there to be a special kind of interaction between the lexical and the prosodic hierarchy. For example, Archangeli &
5The hierarchy could be enriched with more types corresponding to languages, dialects, speakers and registers. For example, certain phonological constraints apply m whole groups of languages (Calder & Bird, 1991 ). Pullcyblank (1989) discuss the tongue mot harmony of Yoruba which is mstrictcd to nouns. If atr was the type of harmonic utter,'mecs, then we could express the necessary constraint hus6: noun=> atr. This kind of constraint is known as a morpheme structure constraint, and phonologists have frequently needed to have recourse to thcse (Kenstowicz & Kissebetlh, 1979, 424fl). A more recent example of the interaction between prosody and morphology is the field of prosodic morphology (McCarthy & Prince, 1990).
A tinal question remains atxmt prosodic constituency. Suppose we wanted to state that a phrase consisted of one or more feet. We could do this using a notation we have already seen, as follows: However, in situations like this where it is easy to write a regular expression, we will simply write: phrase ~ loot ~ where '+' is the Klecne phts. Note that there is a subtle difference of inteq}retation between this notation and a clause of a regular grammar. If we write t ~- ¢ and t ~ ~h, where t is a type and where cp and ~/~ are regular expressions, then both q~ and g, inhabit the t node of the prosodic type hierarchy. We could have written equivalently t -> O rq ~/,. In contrast o this, if t --, 4) and t ~ ¢, are clauses of a regular grammar, t could be rewritten as either 0 or ~,, i.e. i -, 0 kl g,. Now we progress to a more detailed example to illustrate the framework.
"rer&ta is an Arawakan language spnken in Brazil, described by Bendor-Sanuet (1970). It has the li31lowing segntent inventory:
stops: p,t,k, ?
fricatives: s, J', h, t t
liquids: 1, r,m,n semi-vowels: y, w
vowels: i,e, a, o, u
6Itere Iassume that prosodic types atre actually l}mperties of whole signs rather than just the phonology attribute. AcrEs DE COLING-92, NhNqES, 23-28 hOiST 1992 7 7 The categoty of first person is marked by a prosody of nasalisation on both nouns mid verbs, its tim data in Table 1 illustrates. Table 1: Terrena Data e'nm?u his word 6Ptn6?fi my word 'ayo his brother '@6 my brother 'owoku his house '6fv6Ugu my house 'piho he went 'mbiho I went a' qa?aJ'o tie desires ~i'n3a'?aJ'o I desire
Tim seginents of the words in the right column am all nasalised until the lirst obstrucnt (other titan ?) or the end of the word. The ubstntent is prenas,'dised and voiced, but we can ignore the voicing as it is itot eonlrastive in the language,
Art tmalysis of this data m the computational frantework outlined alx)vc might proceed as follows. First, suppose that there is a distinctive feature of nasality which can mark all scgtncms, h sepccs Io create the tollowmg distinction7: Obstruents Snnorants Vowels ~N _+N p rot} t "d k Ug • ' "5 n Z h "3 :L . 1 shall employ capital letters to denote each pair (e.g. P = {p, '"b}, S = {s, "z} and so on). Suppose that N denotes the class ofnasalised segments and O denotes the class of obstnmttts (except htr ?). Ilere is the feature smteture fin' tile lirst person morpheme.
\[ I}II()N (NfI())*((NNO)N')\] , \[IIEAI)IMAJ NIIM\]
SYN'I:K: \[ SUBCAT 0 J s I.;M IC(}NTI INI) I VA~: I PI¢I~, /.~: ~N
r
1 m y w . - +N---W
? i
\]" e
m a
u .9 * . +IN-\] f 6 ;i u t . . The phonology attribute is of most interest here. It specilies that there is a (possibly empty) sequence of nasal tlon-obstnmnts, tolluwed optitmally by a nasal ale \[ RANSPARENT It} the harmony process; Ihey neither ahemate \[lot Mock \[lltrlll(llly. PROC. OF COLING-92, NANTES, AUG. 23-28, 1992 obstruent and a sequence of non-nasals. Notice that with this simple regular expression the notions of autosegruental SPREADING and BLOCKING are captured, This approach will therefore generalise to such phenomena s vowel harmony. Here then is part of the lexical entry for the noun house. PHON 'OWOKU I \[ HFaDIMAJ N SYNILOC | /rSYN\[LOCIHEADIMAJ NUM\]\ S~M\]CONT\[B IIN,IRf~TIREt.N HOUSI': Combining these two feature structures produces the following result: ~t.ION IYNILOC 5~'6Ugu \[11 ~I~DIMAJ N\] \[SUBCAT 0 SEMICONTIIND \[V/~ LRI~TIRELN lPb:RS 1St IIOUSH
It has frequently been observed that there is a prosodic constraint involved in the following data (based on Culicover (1976, 156)): (4) a.
b.
c. d. Mary \[called \[John I \[an elephant\]\] *Mary \[called \[an elephant\] [John\]\] Mary \[called \[the man who stepped on her foot\] \[an elephant\]\] Mary \[called \[an elephant\] [the man who stepped on her foot\]\] In generative syntax this is called complex (or 'heavy ') NP shift, since it involves the movement of a direct object NP past another complement or adjunct, and this movement is only possible if the phrase being moved is sufficiently large. Recall that Pollard and Sag's grammar for English has two linear precedence constraints. The first (LP1) states in effect that heads precede complements. The second (LP2) states that the ordering of complements respects the obliqueness hierarchy. LP2 is repeated below; the symbol '<<' is the obliqueness ordering carried over into phonology (Pollard & Sag, 1987, 174). AcrEs DE COLING-92, NANTES, 23-28 AOI)T 1992 7 8 (5) LP2:
COMPL << COMPL\[LEX -\]
Complements must precede more oblique phrasal
complements. Pollard & Sag (1987, 178) analyze complex NP shift, claiming that the' moved' constituent isfocussed and that focussed constituents can violate LP2. They further claim that it is not possible to 'shift' the constituent past an NP sister. This leads to the constraint \[MAJ -N\] < \[FOCUS +\], which must be disjoined with the original LP2 as shown in (6), a technique due to Uszkoreit (1986). (6) COMPL ,(< COMPL\[LEX -\]
V \[MAJ -N\] .< \[FOCUS +\] This constraint conforms to more general observations that focussed material tends to appear sentence finally (Halliday, 1967). However, observe that in (4d), the 'shifted-over' constituent is in fact an N~', contrary to the \[MAI -N\] requirement of (6) 8. The LP constraint must be revised as follows. (7) LP2 (first revision): COMPI.\[FOCUS -\] <~<~ COMPL\[LFL~ -\] Unfocussed complements must precede more oblique phrasal complements.
FOCUS is clearly not the only factor involved in complex NP shift. Sentence (4b) is bad regardless of whether or not john is focussed. An interesting attempt o unravel the prosodic contribution to this phenomenon has been made by Zec & lnkelas (1990, 376-7). According to them, a prosodic constituent is said to be ttEAVY just in case it branches. This is familiar in the case of syllables: a heavy syllable is one having two (or more) moms, while a light syllable only has one (Hyman, 1984). Rather then presenting the details of their analysis, I shall present a liberal adaptation of it that is more suited to the phonological framework assumed here. Suppose that an intonational phrase consists of a sequence of stress feet 9, feet consist of syllables and syllables consist of segments. Using the notation described in section 2, we could write:
8Culicover (1976, 155-6) gives more examples to support this claim. Other verbs allowing movement past an NP are the L)I verbs of (Quirk et al.. 1972, 850). Dative verbs do not permit complex NP shift.
9Whether there is intervening material between the phrase and the foot is immaterial tothe present discussion. PROC. OF COLING-92, NANTES, AUG. 23-28, 1992 phrase =t, foot + foot : ~ syl + syl : =¢- kal t_l step tJ fot U 6a t_l •.. To this we add some further constraints. First, heavy phrases (phraseh) must contain two or more feet. Second, feet consist of exactly one heavy syllable, while all other syllables must be light. Third, the vowel of a light syllable must be a schwa. phrase h =~ foot foot + foot ~ ,vyl~ ,~yl h .vyl~ syl I ::¢, C* ~ C" C ~ bUdtJ6 H fLIgt_l ... Although these specifications are admittedly rather crude, they are sufficient for present purposes. The next step is to set up an abbreviation to enable heavy and light phrases to be picked out easily. t IEAVY =\[PHONphraseh\] LIGIIT ~ \[PHON phraset\] Now we can formulate an ordering principle that accounts for the observed behaviour. This is done by modifying LP2. (8) LP2 (final version):
COMPL\[LIG|tT\] <(( COMPI,\[LEX -1
Light complements mu.vt precede more oblique
phrasal complements. Clearly, a combination of (7) and (8) is needed if Pollard and Sag's observations are to be linked with those of Zec and lnkelas. However, such a relinement would still be inadequate. The acceptability of shifting of an NP actually gets worse as the size of the material between the verb and the NP gets larger. Thus the relative weight of the complements i important. It is apparent hat the spectrum of relative weights interacts with the obliqueness ordering in a gradient manner that cannot be adequately dealt with in the polarised fashion of linear precedence constraints.
In the course of this paper Ihave presented aphonological framework and shown how it might be interfaced with IIPSG. The frmnework starts where HPSG Aeries DF. COLING-92, NANTES, 23-28 Ao(rr 1992 7 9 left off, in the sense that \]IPSG's phonology attribute-a list--can be viewed as a subclass of automata. The framework is potentially far-reaching in the sen~ that it encompasses non-linear phonology (Bird & Ellison, 1992). Although this work is preliminary, it is alleady clear how several standard phonological devices may be recast in this new framework. For exanrple, the traditional distinction between morphophonology and phonology proper and the distinction between lexical and post-lexical phonology can be modelled in terms of the distinction between lexical and prosodic hierarchies. The traditional prosodic domains (phrase, foot, syllable) are recast as types, and domain-bounded processes are regular expressions tied to the required prosodic type.
One area of morphophonology that has not been touched on here is the situation where the selection of an allomorph is sensitive to the phonological context. A well-known example of this is the a~an altemation in English. One solution would be to represent contextual inlbm~ation as part of the phonology attribute of a sign by associating prosodic types with final states. c ::~ C.* v~-V.* When this morpheme is concatenated with a following noun, these prosodic types must be brought to bear on the following material. Another area for i'urther work is to investigate interactions between phonology and semantics. For example, it is often observed that there is a close correspondence between intonational phrases and a particular semantic luster called a sense unit, whereby any stretch of material that correspol:tds to an intonational phrase must also qualify as a sense unit (Bird, 1991b). Previous accounts of this phenomenon have required the mediation of a syntactic module and this has lead to an extra level of complexity (e.g. Selkirk (1984, 290-296)). Now there is the exciting prospect of being able to explore phonology-semantics nteractions directly. For example, given suitable definitions, we could simply writc: PIIRASE ~- SI,'.NSE-UNIT. It is hoped that further work in this vein will lead to a nonlinear model of phonology that is fully integrated into constraint-based grammar formalisms. This will open the way for applications of such frameworks to a much wider range of the world's languages. PRoc. ov COLING-92. NAr'VrES, AUG. 23-28. 1992 References Antworth, E. (1990). PC-KIMMO : A Two-Level Processor
forMorphologicalAnalysis. Dallas: SIL. Archangeli, D. & Pulleyblank, D. (1989). Yoruba vowel
harmony. Linguistic Inquiry, 20, 173-217. Bach, E. (1983). On the Relationship between Word-
Grammar and Phrase-Granlmar. Natural Language
and Linguistic Theory, 1(1), 65-89. Bendor-Samuel, J. T. (1970). Some problems of segmen-
tation in the phonological analysis of Terena. In
F. R. Palmer lEd.), Prosodic Analysis chapter 14,
(pp. 214-221). OUP Bird, S. (1990). Constraint-BasedPhonology. PhD the-
sis, University of Edinbnrgh. \[To be published by
Kluwer\]. Bird, S. lEd.). (1991a). Declarative Perspectives in
Phonology, volume 7 of Working Papers in Cog-
nitive Science. Bird, S. (1991b). Focus and phrasing in Unification Cat-
egorial Grammar. In S. Bird lEd.), Declarative Per-
spectives on Phonology (pp, 139-166), University of
Edinburgh. Bird, S. & Ellison, T. M. (1992). One level phonology: an-
tosegmentai representations and rules as finite-state
automata. RP 51, University of Edinburgh, Centre
for Cognitive Science. Bird, S. & Klein, E. (1990). Pbonological events..lournal
of Linguistics, 26, 33 56. Bird, S. & Ladd, D. R. (1991). Presenting autoscgmental
phonology. Journal of Linguistics, 27, 193-210. Calder, J. & Bird, S. (199l). Defaults in nnderspecifica-
tion phonology. In S. Bird lEd.), Declarative Per-
spectives on Phonology (pp. 107-125). University of
Edinburgh. Coleman, J. S. (1991). PhonologicalRepresentations
their names,forms and powers. PhD thesis, Univer-
sity of York. Culicover, P. W. (1976). Syntax. Academic Press. Dowty, D. R., Wall, R. E., & Peters, S. ( 1981 ). Introduction
to Montague Semantics. Reidel. Goldsmith, J. (1976). Autosegmental Phonology. PhD
thesis, MIT. \[Published New York: Garland, 19791. Halliday, M. A. K. (1967). Intonation and Grammar in
British English. Mouton. Hooper, J. (1976). An Introduction to Natural Generative
Phonology. Academic Press. Hyman, L. M. (1984). On the weightlessness ofsyllable
onsets. In Proceedings of the Tenth Annual Meeting
of the Berkeley Linguistics Society. Acq'I,:s DE COLING-92, NA>rI'gS, 23 28 ao~r 1992 8 0 Johnson, M. (1988). Attribute-Value Logic and tile Theory
of Grammar. PhD thesis, Stanford Uuiversity. \[CSLI
Lecture Notes Volume 16\], Kay, M. (1987). Nonconcatenative tinite-state morphol-
ogy. Proceedings of the Third Meeting of the Euro-
pean Chapter of the Association for Computational
Linguistics, 2-10. Kenstowicz, M. & Kisseberth, C. (1979). Generative
Phonology: Description and Theory. Academic
Press. Kornai, A. (1991). Formal Phonology, PhD thesis, Stan-
ford University. Koskenniemi, K. (1983). Two-Level Morphology: A Gen-
eral Computational Model for Word-Form Recog-
nition and Production. PhD thesis, University of
Hclsinki. McCarthy, J. & Prince, A. (1990). Four and word in
prosodic morphology: tile Arabic broken plural. Nat-
ural Language and Lingasitic Theory, 8(2), 209-
283. Partee, B. H. (1979). Montague grammar and the well-
formedness constraint. In F. Heny & H. Schnelle
(Eds.), Syntax and Semantics 10: Selections from
the Third Groningen Round Table (pp. 275-313).
Academic Press. Pierrehumbert, J. (1990). Phonological and phonetic rep-
resentation. Journal of Phonetics, 18, 375-394. Pollard, C. & Sag, I. (1987). Information-Based Syntax
and Semantics, volume 13 of CSLI lecture notes.
CSLI. Quirk, R., Grecnbaum, S., Leech, G., & Svartvik, J.
(1972). A G rummar of Contemporary English. I,ong-
man, Rice, K. (1989). On eliminating resyllubification i to on-
sets. Proceedings of the West Coast Conference on
Formal Linguistics, 8, 331-346. Scobbie, J. (1991 ). Attribute- Value Phonology. PhD thesis,
University of Edinburgh. Sclkirk, E. (1984), PhonologyandSyntax. The M1T Press. Shieber, S. (1986). An Introduction to Unification-Based
Approaches to Grammar. Stanford:CSLl. Uszkoreit, H. (1986). Constraints on order. CSLI 86-46,
Stanford University, Wheeler, D. (1981). Aspects of a Categorial Theory of
Phonology. PbD thesis, U.Mass, Amherst. Zec, D. & lnkelas. S. (1990). Prosodicully constrained syn-
tax. In S. lnkelas & D. Zec (Eds.), The Phonology-
Syntax Connection chapter 18, (pp. 365-378). The
University of Cbicago Press, PRO(:. ot: COLING-92, NANTES, AUG. 23-28, 1992