hierarchical structure in the subsegmental area

Kaye, Jonathan, Jean Lowenstamm & Jean-Roger Vergnaud 1985. The internal structure of phonological representations: a theory of Charm and Government.
163KB taille 0 téléchargements 327 vues
Tobias Scheer CNRS 6039, Université de Nice [email protected] this handout and more stuff at www.unice.fr/dsl/tobias.htm

Les géométries de traits : de la phonologie à la syntaxe Paris 3 December 2005

HIERARCHICAL STRUCTURE IN THE SUBSEGMENTAL AREA: WHY AND HOW (1)

purpose a. answer the question "are all features hierarchically equal in phonology ?"

b. c.

d. (2)

so that syntacticians are informed when they, eventually, are tempted to import phonological reasoning regarding the hierarchical organisation of primes. the answer to the question is NO there is probably no phonologist on Earth who would answer YES. ==> so the real purpose is to explain why everybody answers NO. I could identify 3 reasons: 1. natural classes 2. inclusive relationships under processing 3. unequal contribution to the result plus theory-internal reasons, rather unhelpful: 4. expressive power of the grammar and systemic properties of sound 5. the general dependency philosophy: "nothing is equal in linguistics" in order to see that, it is useful to first look at HOW hierarchical relations have been formalised: SPE > binary vs. monovalent approaches.

historical excursus: melodic representation in phonology a. neogrammarians talked about "labiality", "obstruents" etc. all the time, but there was no attempt to formally or even only graphically represent these units. b. the same holds true for Prague structuralists: a phoneme may be different from another phoneme by its labial character, so there is a labial correlation, but the relevant categories remained words - there was no formal or graphic representation either. c. Jakobsonian binarism in the 50s Jakobson introduced the binary idea: [± α], where α had an acoustic definition: e.g. [± strident]. d. SPE took over binarity, but made the labels articulatory-based: [± labial]. "feature bundle": a feature matrix was an amorphous, unordered and nonhierarchiesed list of binary features. e. as the autosegmental idea spread in the late 70s - early 80s, all areas of phonology were transformed accordingly: 1. subsegmental structure: Feature Geometry (Clements 1985 et passim). 2. syllable structure (Kahn 1976 et passim). 3. representation of suprasegmentals: stress and rhythm (Liberman & Prince 1977 et passim), tone (Williams 1976, Goldsmith 1976 et passim). 4. the interface with higher modules (morphology, syntax) (Selkirk 1981 [1978] et passim).

-2f.

e.

(3)

the monovalent idea (also called privative, unary, holistic) : contrast is not expressed by a binary opposition between positive and negative values of the same object, but by the presence or the absence of the object in question. Hence the difference between [i] and [e] is not binary: [i] = [+high] [-low] vs. [e] = [-high] [-low] but monovalent: [i] = I vs. [e] = I + A I = high front tongue body position, A = low tongue body position. ==> you are present only if your essence contributes to the result. monovalency was originally proposed by Anderson & Jones (1974); it was later implemented by three theories: Dependency Phonology (Anderson & Ewen 1987 et passim), Particle Phonology (Schane 1984 et passim) and Government Phonology (Kaye 1985 et passim). OT is a theory of constraint interaction, not of constraints. As in most other areas, it has therefore simply taken over the representations inherited from of the 80s and adapted them to the new constraint-based environment.

how does this work? Feature Geometry a. here is a typical feature geometric tree A

B C a

b c

D

d

e f b.

c.

g

what are the issues? 1. how many features are there? 2. what are their labels? 3. what are the labels of the non-terminal nodes? 4. are terminal nodes different from non-terminal nodes? 5. is there just one universal tree, or is there cross-linguistic variation? 6. geometric properties of the tree: how are the features put up, which nonterminal dominates which terminal? essentially point 6 is of interest for us. There are two ways of determining arboreal structure 1. natural classes 2. including relationships

-3(4)

natural classes a natural class is a set of objects that behave alike in nature. At two levels: a. static = phonetic/ articulatory examples of natural classes in a 7 vowel system y

i ø

u o

e a

b.

(5)

- high vowels = y,i,u - front vowels = i,e,y,ø - rounded vowels = y,ø,u,o under phonological processing ideally, these are the same as the ones determined by phonetic properties. But in fact on many occasions this is not true: 1. Tiberian Hebrew: r and gutturals cannot geminate 2. although phonetically paired, s,z are not the fricatives corresponding to t,d. Spirantisation invariably shows that T,D are paired with t,d. 3. place of consonants: arguments for all possible groupings exist: labial + coronal coronal + velar = lingual node labial + velar = peripheral node

including relationships = evidence for hierarchical grouping from phonological processes a. in an autosegmental environment, phonological processes consist of linking and delinking of melodic primes. b. it is not true that phonological processes manipulate any two primes. Some primes move together, others never do: - all place features are moved as a block: nasal assimilation i[m]-possible i[n]- terminable i[ŋ]-credible - laryngeal features are moved together: assimilation of aspiration and voicing, e.g. in Classical Greek and Sanskrit: *φτ, *κθ, only φθ, χθ. - single features are moved: vowel harmony (height, roundedness, frontness). - there is no known process by which, say, a velar consonant is palatalised and at the same time voiced (while non-velars remain unvoiced). c. hence the assumption that phonological rules perform single operations only (Clements & Hume 1995:250)

-4d.

and hence in the above geometry A

B C a

b c

D

d

e f

g

1. two or more terminals can only be addressed by a phonological process if there is a node that dominates them 2. it is impossible for a process to address a node without also addressing everything that it dominates. example: - a rule may simultaneously affect d,e,f,g by addressing node C. - no rule may simultaneously affect a,b and e without also affecting b,c,d,f,g. ==> phonological processes operate on terminals or nodes. (6)

natural classes and including relationships are not the same a. natural classes tell us which features are grouped together. b. including relationships tell us how the groups are arranged hierarchically

(7)

how does this work? Monovalent systems a. there are three basic monovalent primes (in upper case latters) that correspond to the extremes of the vocalic triangle: I

U A

b.

I = frontness U = roundedness A = lowness

all other vowels are a combination of those [and also, with a few more primes, consonants and special properties of vowels such as nasality, ATRness etc., but this leads us too far afield here] [i] = I [u] = U [a] = A [e] = I-A [o] = U-A [y] = I-U [ø] = I-U-A

-5c.

d. e.

how are natural classes expressed? - high vowels = absence of A - front vowels = presence of I - rounded vowels = presence of U is there a hierarchical structure among monovalent primes: yes. a dependency relation: one item, the head, is more prominent than the other(s). why should there be such a relation? 1. unequal contribution to the result because the relative prominence of primes in the result is not the same: "old" version: it expresses the phonetic result (heads underscored) I-A = [ɛ] - closer to I because I is head I-A = [Q] - closer to A because A is head ==> the head contributes more to the phonetic result than the operator. 2. a more modern, non-phonetic version of this idea: application to sonority, which must not be a prime, i.e. there is no [± son] prime. Sonority is a derived category. Several proposals along these lines: Rice (1992), Scheer (1999,2004:§46), Hermans (2003). Scheer (1999): sonority is the result of the interplay of three factors: - whether the primes are attached to an Onset or a Nucleus - whether the laryngeal primes h and ʔ are present or not - whether the maximum-sonority prime A is absent, operator or head segment a e,o i,u liquids nasals glides s,z gutturals fricatives stops

Nucleus or Onset N N N O O O O O O O

h/? — — — — — — h h h h and ?

role of A head operator absent head head/ operator absent head head/ operator operator/ absent absent

sonority

-63. in order to be able to express natural classes in a perspective that incorporates the fundamental structuralist insight: segments are also subjected to systemic properties of sound. Otherwise the expressive power of the grammar is not big enough. Licensing Constraints: within a segmental expression, the head licenses the operator(s). A particular system is the result of restrictions that constraints on this Licensing impose. E.g. - operators must be licensed - A is not a licensor - U must be head ==> a given phonetic object may have very different phonological identities in different systems according to its relations with the other members of the system, e.g. [ɛ] = I-A or I-A or I-A headless Literature on Licensing Constraints: Charette & Göksel (1994,1996), Cobb (1997), Kaye (2001), for a summary: Scheer (2003). 4. because of the general Dependency philosophy: all linguistic structure is a matter of dependency: Anderson (1987, 1992). (8)

comparison binary - monovalent a. the hierarchical relationship among primes is expressed by - trees (binary) - head-operator (monovalent) b. an argument for each (classical) 1. high vowels cannot be referred to as a natural class in a positive fashion: the only thing that unites them is the absence of A. 2. something that isn't there cannot spread: correct prediction for nasality by monovalent systems since [+ nasal] is often observed to spread, but [- nasal] never does: there is no rule of the kind "a nasal vowel becomes non-nasal in the presence of a non-nasal segment". Binary systems, however, predict the spreading of [- nasal].

(9)

hybrid systems "tree cum monovalency": underspecification a. underspecification is the import of monovalency into a binary and arboreal environment. b. underspecification is anti-redundancy: the lexicon must not contain any predictable information (SPE). An underspecified feature is one that is absent at a certain level of representation (lexically and "early on" in the derivation) and joins in only later (by default or fillin rules). Hence while it isn't there it cannot spread.

-7c.

argument against underspecification: it is worse than the binary system because it creates a three-way opposition: a feature may 1. have positive value 2. have negative value 3. be absent

(10) hybrid systems "monovalency cum tree" a. some proponents of monovalency have arranged monovalent primes on a geometric tree: Weijer (1994), Harris (1994) b. but not with original arguments: the regular geometry known from binary systems is just copied, and no specific arguments are provided: Laryngeal separated from Place etc. c. Harris' (1994:129) system: x Root ʔ Laryngeal

N

h

Place A I

U

R

(11) markedness comes in with underspecification a. if underspecification is accepted, how do we decide which features are underspecified? Two philosophies: 1. Radical Underspecification unmarked features are underspecified [e.g. Kiparsky 1982, Archangeli 1988, Pulleyblank 1988, Paradis & Prunet 1989] 2. Contrastive Underspecification non-contrastive features are underspecified [e.g. Steriade 1987, Clements 1987, Mester & Itô 1989] 3. a "best of two worlds" synthesis is argued for by Avery & Rice (1989): Modified Contrastive Underspecification: a universal markedness theory provides default information as to which features are present/ absent in underlying representations - this general picture is then modified by existing contrasts in a given language: for each contrast, only the marked feature will be present lexically. b. differences 1. Radical Underspecification underspecification is universal: it is determined by a universal theory of markedness. Hence there is only one universal feature geometric tree.

-8-

c.

2. Contrastive Underspecification underspecification is language-specific: whether a feature is underspecified has got nothing to do with markedness, but depends on the phonemic contrasts found in every language. A structuralist piece in the theory. Hence feature geometric trees are not the same in all languages at the terminal level. interminable debates on (universal) markedness: for place, coronal is favourite, but there are also good arguments for the velar locus. [e.g. Paradis & Prunet 1991, Rice 1994,1996,1999a,b

(12) does the underspecification debate affect the geometric properties of the tree ? a. No: the tree is virtually still the same in all languages: possible variation merely concerns the instantiation of a sub-parts of it: if the terminal element is absent, the "empty" structure that dominates this element is usually also taken to be absent. Reason: locality, cf. below. The geometry itself is still the same in all languages. b. in some versions, however, (Modified Contrastive Underspecification), even "empty" arboreal structure is present at all levels. (13) arguments in favour of underspecification a. 1. elimination of redundancy - old 2. what isn't present can't spread - old 3. locality - new b. locality (= transparency to spreading) 1. absolute autosegmental law: non line-crossing 2. what do we do with facts where lines do seem to cross (= transparency) ? c.

Latin -alis / -aris allomorphy (e.g. Roca 1994:54ss) I II III IV nav-alis sol-aris flor-alis regul-aris voc-alis milit-aris litor-alis articul-aris 1. 2. 3. 4.

-alis is underlying, cf. I. dissimilation to -aris if the root contains a lateral, cf. II but no dissimilation if there is an intervening rhotic, cf. III a radical rhotic before the lateral does not block dissimilation, cf. IV.

d. x | s

x | f

x | o

x x x x | | | | | a | i [+lat] [+lat] dissimilation

x x | | | o | [+lat]

x | | | [-lat]

x | a

no dissmilation

x x | | | i | [+lat]

x | s

x | s

dissimilation goes "through" a vowel, and "through" consonants: milit-aris.. ==> vowels and non-liquids are invisible.

dissimilation is blocked by another liquid ==> liquids are visible

-9-

e.

− if all segments are specified for [± lateral], there could not be any transparent segments without line-crossing. − obviously [± lateral] is only used to distinguish between r and l, all other segments are [- lateral] by default. − hence underspecifying [± lateral] for its negative value produces two desirable results: 1. the whole operation is local: at the level of the feature [± lateral]. 2. transparency follows: there is no line-crossing.

(14) summary binary vs. monovalent primes (leading to markedness and underspecification) a. no incidence on our purpose: 1. the tree, if any, will always be the same 2. no new arguments for determining its geometric properites b. rather than empirical, the monovalent side has theory-internal arguments for proposing a hierarchical relationship among primes: 1. the very general "there are no peers in linguistics, all domains are hierarchically organised" 2. further generative power is needed in order to express the existing contrasts. 3. heads are prominent: the more you contribute the more prominent will your essence be in the result. (15) feature theory in OT a. OT is a theory of constraint interaction, not of constraints. As in most other areas, it has therefore simply taken over the representations inherited from of the 80s and adapted them to the new constraint-based environment. b. OT thus is a computational theory, and any "object" could do as an input. Hence a natural decline in interest regarding representations: they are taken over from the 80s by default, but they do not serve any purpose. There is no agrammaticality anymore, and hence line-crossing for example is toothless. OT has no trouble at all in expressing a situation where a line-crossing candidate wins because its competitors do not cross lines, but are worse on some other count. c. There are explicit statements regarding the arbitrary interchangeability of melodic representations: "The tenets of OT, regarding constraint violability and ranking, make no particular claims about phonological representations. We could, for example, do OT with any kind of feature theory: SPE feature bundles or feature geometric representations, privative or binary features, and so on." Lombardi (2001:3)

- 10 d.

Hence there is a strong loss of interest in the debate regarding the nature of primes (binary vs. monovalent), and also regarding their hierarchical organization. "One less desirable consequence [of the rise of constraint-based phonological theories] has been an increasing uncertainty regarding such fundamental questions as: What is a lexical representation? What is a phonological representation? Of what features or feature specifications do they consist? How do these features combine? What is the trade-off between constraints and representations in understanding phonological regularities?" Clements (2001:71) "A consequence of the shift away from representational questions […] is that there is at present much uncertainty concerning certain fundamental questions pertaining to […] phonological representations […]. With respect to features, the most obvious question […] is: What featural representations (e.g. feature geometry, underspecification) are necessary in a phonological theory?" Hall (2001:1)

e.

f. g.

if representations and the hierarchical organisation of melodic primes are any relevant at all in OT, something which one can doubt, there is probably no contribution coming from this side regarding − the geometric properties of the tree and − the motivation for the existence of hierarchical structure. looking at current practice in OT, analysis of melody largely reduces to constraints of the type *X: *DORSAL, *LABIAL, *[+cont], hence without any reference to tree structure or any specific geometric organization. also, it may well be the case that the angle stone of Feature Geometry, i.e. the assumption that there is only one universal tree, does not frame melodic representations anymore. "feature organization is universally determined" Clements & Hume (1995:250)

(16) conclusion a. operationally, if you want to import phonological reasoning into syntax, you will probably have to look for 1. natural classes 2. inclusive relationships under processing 3. unequal contribution of syntactic features to the result. b. the theory-internal reasons mentioned are probably not very helpful: 4. expressive power of the grammar and systemic properties of sound 5. the general dependency philosophy: "nothing is equal in linguistics" c. beyond the hierarchical organisation of syntactic features, you may also want to consider their nature: binary, monovalent, underspecified (hence touching on markedness). d. as you will have noticed, feature geometric trees do not obey any formal requirement: 1. nothing like X-bar 2. nodes can branch as many times as the linguist wants them to 3. there are no designated formal categories such as complement, Spec, X0, X', X'' etc.

- 11 References References followed by "WEB" are available at www.unice.fr/dsl/tobias.htm. Anderson, John 1987. Structural Analogy and Dependency Phonology. Explorations in Dependency Phonology, edited by John Anderson & Jacques Durand, 15-48. Dordrecht: Foris. Anderson, John 1992. Linguistic representation: Structural analogy and stratification. Berlin & New York: Mouton de Gruyter. Anderson, John & Colin Ewen 1987. Principles of Dependency Phonology. Cambridge: Cambridge University Press. Anderson, John & Charles Jones 1974. Three theses concerning phonological representations. Journal of Linguistics 10, 1-26. Archangeli, Diana 1988. Aspects of underspecification theory. Phonology 5, 183-208. Avery, Peter & Keren Rice 1989. Segment structure and coronal underspecification. Phonology 6, 179-200. Charette, Monik & Asli Göksel 1994. Vowel Harmony and Switching in Turkic languages. SOAS Working Papers in Linguistics and Phonetics. 4, 31-52. Also in Kardela, Henryk, Bogdan Szymanek (eds), A Festschrift for Edmund Gussmann, 29-56. Lublin 1996: University Press of the Catholic University of Lublin. WEB. Charette, Monik & Asli Göksel 1996. Licensing constraints and vowel harmony in Turkic languages. SOAS Working Papers in Linguistics and Phonetics 6, 1-25. Also in Cyran, Eugeniusz (ed), Structure and Interpretation. Studies in Phonology, 65-88. Lublin 1998: Folium. WEB. Clements, George 1985. The geometry of phonological features. Phonology 2, 225-252. Clements, George 1987. Phonological Feature Representation and the Description of Intrusive Stops. Chicago Linguistics Society 23, 29-50. Clements, George 2001. Representational economy in constraint-based phonology. Distinctive Feature Theory, edited by Tracy Hall, 71-146. Berlin, New York: Mouton de Gruyter. Clements, George & Elizabeth Hume 1995. The Internal Organization of Speech Sounds. The Handbook of Phonological Theory, edited by John Goldsmith, 245-306. Oxford: Blackwell. Cobb, Margaret 1997. Conditions on Nuclear Expressions in Phonology. Ph.D dissertation, SAOS, London. Goldsmith, John 1976. Autosegmental Phonology. Ph.D dissertation, MIT. Hall, Tracy 2001. Introduction: Phonological representations and phonetic implementation of distinctive features. Distinctive Feature Theory, edited by Tracy Hall, 1-40. Berlin, New York: Mouton de Gruyter. Harris, John 1994. English sound structure. Oxford: Blackwell. Hermans, Ben 2003. Russian Vowel Reduction as Diminution of A-Complexity. Talk presented at Formal Description of Slavic Languages 5, Leipzig, November 1996. Kahn, Daniel 1976. Syllable-based generalizations in English phonology. Ph.D dissertation, MIT. Published by Garland Press, New York 1980. Kaye, Jonathan 2001. Working with licensing constraints. Constraints and Preferences, edited by Katarzyna Dziubalska-Kołaczyk, 251-268. Berlin & New York: Mouton de Gruyter. WEB. Kaye, Jonathan, Jean Lowenstamm & Jean-Roger Vergnaud 1985. The internal structure of phonological representations: a theory of Charm and Government. Phonology Yearbook 2, 305-328. WEB. Kiparsky, Paul 1982. From Cyclic Phonology to Lexical Phonology. The structure of

- 12 phonological representations I, edited by Harry van der Hulst & Norval Smith, 131175. Dordrecht: Foris. Liberman, Mark & Alan Prince 1977. On Stress and Linguistic Rhythm. Linguistic Inquiry 8, 249-336. Lombardi, Linda 2001. Introduction. Segmental phonology in Optimality Theory. Constraints and Representations, edited by Linda Lombardi, 1-9. Cambridge: Cambridge University Press. Mester, Armin & Junko Itô 1989. Feature predictablilty and underspecification: palatal prosody in Japanese mimetics. Language 65, 258-293. Paradis, Carole & Jean-François Prunet 1989. On coronal transparency. Phonology 6, 317348. Paradis, Carole & Jean-François Prunet (eds) 1991. The Special Status of Coronals: Internal and External Evidence. San Diego: Academic Press. Pulleyblank, Douglas 1988. Underspecification, the feature hierarchy and Tiv vowels. Phonology 5, 299-326. Rice, Keren 1992. On deriving sonority: a structural account of sonority relationships. Phonology 9, 61-99. Rice, Keren 1994. Peripheral in Consonants. Canadian Journal of Linguistics 39, 191-216. Rice, Keren 1996. Default variability: the coronal-velar relationship. Natural Language and Linguistic Theory 14, 493-543. Rice, Keren 1999. Featural markedness in phonology: Variation. Part 1. Glot International 4.7, 3-6. Rice, Keren 1999. Featural markedness in phonology: Variation. Part 2. Glot International 4.8, 3-7. Roca, Iggy 1994. Generative Phonology. London: Routledge. Schane, Sanford 1984. The fundamentals of particle phonology. Phonology Yearbook 1, 129155. Scheer, Tobias 1999. A theory of consonantal interaction. Folia Linguistica 32, 201-237. WEB. Scheer, Tobias 2003. Privative melodic representations and structuralist thinking: Licensing Constraints. Paper presented at Colloque du PFC, Paris 11-13 décembre 2003. WEB. Scheer, Tobias 2004. A Lateral Theory of Phonology. Vol.1: What is CVCV, and why should it be? Berlin: Mouton de Gruyter. WEB. Selkirk, Elisabeth 1981 [1978]. On prosodic structure and its relation to syntactic structure. Nordic Prosody II, edited by Thorstein Fretheim, 111-140. Trondheim: TAPIR. Steriade, Donca 1987. Locality Conditions and Feature Geometry. Proceedings of NELS 17, edited by Joyce McDonough & Bernadette Plunkett, 595-617. Amherst: GLSA. Weijer, Jeroen van de 1994. Segmental structure and complex segments. Ph.D dissertation, University of Leiden. Williams, Edwin 1976. Underlying tone in Margi and Igbo. Linguistic Inquiry 7, 463-484.