




























































































Study with the several resources on Docsity
Earn points by helping other students or get them with a premium plan
Prepare for your exams
Study with the several resources on Docsity
Earn points to download
Earn points by helping other students or get them with a premium plan
Community
Ask the community for help and clear up your study doubts
Discover the best universities in your country according to Docsity users
Free resources
Download our free guides on studying techniques, anxiety management strategies, and thesis advice from Docsity tutors
Material Type: Notes; Class: Phonology I; Subject: Linguistics; University: University of Delaware; Term: Spring 2008;
Typology: Study notes
1 / 165
This page cannot be seen from the preview
Don't miss anything!
On special offer
Acknowledgements
These notes draw heavily on the notes of Colin Wilson’s Fall 2002 and especially on Kie Zuraw’s Fall 2005 UCLA graduate phonology classes. Although I have made some changes, in many cases, they are word-for-word the same. There are undoubtedly many typos. Please email them to me as you find them. -Jeff Heinz (Feb 12, 2008)
Language learning would be impossible unless this were the case. The important question is: What are the initial assumptions con- cerning the nature of language that the child brings to language learning, and how detailed and specific is the innate schema (the general definition of “grammar”) that gradually becomes more ex- plicit and differentiated as the child learns the language? For the present we cannot come at all close to making a hypothesis about innate schema that is rich, detailed, and specific enough to ac- count for the fact of language acquisition. Consequently, the main task of linguistic theory must be to develop an account of linguis- tic universals that, on the one hand, will not be falsified by the actual diversity of languages and, on the other, will be sufficiently rich and explicit to account for the rapidity and uniformity of lan- guage learning, and the remarkable complexity and range of the generative grammars that are the product of language learning. (Chomsky 1965, Aspects, 27; on explanatory adequacy see Chom- sky 1965: 30-32, 34-37, ch 1, etc.)
(3) Understand how knowledge is applied in particular behaviors, and what limitations prevent performance from being coextensive with competence. [performance theory] There seems to be little reason to question the traditional view that investigation of performance will proceed only so far as un- derstanding of underlying competence permits.... To my knowl- edge, the only concrete results that have been achieved and the only clear suggestions that have been put forth concerning the theory of performance, outside of phonetics, have come from studies of per- formance models that incorporate generative grammars of specific kinds.... (Chomsky 1965, Aspects, 10, in ch 1.2: Toward A Theory of Performance)
1.2 Generative phonology
(4) Knowledge of linguistic sound patterns ← we’ll be mainly talking about this a. Representations features, feature geometry, autosegmental representations; natural-class metrics; maps of articulatory difficulty and perceptual similarity;.... b. Generalizations
“linguistically significant generalization”; rules, rule ordering, rule iter- ation, cyclic application; constraints, constraint conflict; status of ex- ceptions;....
(5) Learning sound patterns
a. Partial theories with significant innate components (e.g., Ellison (1992, 1994b), Gildea and Jurafsky (1996), Tesar and Smolensky (2000), Al- bright and Hayes (2003), Hayes and Wilson (2008), Heinz (2007)) b. Largely negative (or simply absent) results concerning the computa- tional feasability of the theories that have been proposed (e.g., Eisner (2000), Idsardi (2006))
(6) Phonological performance — considered mostly as evidence for theories
a. Traditional sources: descriptive grammars, fieldwork, phonetic experi- ments (also poetry, verse, etc.) b. Psycholinguistics: ‘wug’-testing and rating novel forms for morphologi- cal and/or phonological well-formedness (e.g. (Albright and Hayes 2003, Zuraw 2000)); induced speech errors (e.g. (Goldrick 2002)) c. Child phonology: increasingly detailed analyses, largely in response to development of computational learning theories (e.g. (Tessier 2006)). d. Brain-imaging: studies have investigated localization of natural-class / phonological-feature representations (e.g. (Phillips et al. 1995)) e. Relatively little work in the opposite direction — incorporating phono- logical competence into theories of speech production/perception, word recognition, and sentence parsing.
(7) Course issues
a. What methods can be used to define “linguistically significant general- ization”? And how does ‘generalization’ in phonology relate to ‘gener- alization’ in psychology? b. In what form are individual phonological generalizations internalized by the native speaker? rules, morpheme structure conditions, constraints, filters, protototypes,... c. How do generalizations about phonological structure interact with each other? rule-ordering vs. constraint-ranking vs. mixed rule/constraint theories; cyclic interaction d. What innate restrictions or biases are imposed on phonological gener- alizations? And how do these biases relate to the phonetic properties of sounds and sound sequences? markedness, naturalness, formal uni- versals
to expect that any particular insight is going to take us a very small fraction of the way to the kind of truly intelligent mechanisms we seek. The only way to reasonably expect to make progress is by chaining to- gether many such small steps. And the only way to chain together these steps is to understand at the end of each one where we are, how we got there, and why we got no further, so we can make an informed guess as to how to take the next small step. A ‘magic’ step is apt to be a last step; it is fine, as long as it takes you exactly where you want to go. (emphasis original) (Smolensky, Paul. 1986. Information Processing in Dynamical Sys- tems: Foundations of Harmony Theory. In David E. Rumelhart, James L. McClelland, and the PDP Research Group, Parallel Distributed Pro- cessing, Vol. 1: Foundations. Cambridge, Mass.: MIT Press, p.220)
(9) Theories as composite entities
a. Every serious theory of phonology has many, many components. (i) SPE: rule notation, ordering, morpheme structure rules, linking, exception features,... (ii) OT: richness of the base, universal constraints, strict domination, factorial typology;... b. Goal of theory evaluation is therefore very rarely wholesale endorsement or rejection. c. Focus instead on separating chaff from wheat—on identifying the com- ponents that lead to specific incorrect predictions, and on replacing them with better alternatives. Impossible without first understanding the individual components and their interactions.
(10) Suspension of commitment and disbelief
a. Do not commit fully to any particular theory, nor reject theories on general grounds. b. Focus instead on working out the specific predictions of any given theory, looking for new connections among facts as well as internal contradictions and falsifications.
1.3 Generalization in phonology
(11) Knowledge about the sounds and sound sequences that are permitted in surface forms. a. (English) All syllable- or word- final NC clusters are homorganic.
b. (English) All morpheme-internal obstruent clusters are voiceless (cf. adze, rugby, asbestos). c. (Russian) Voiced affricates (e.g. [Ã]) and voiced velar fricative ([G]) occur only as the result of voicing assimilation (Halle 1959). d. (Malayalam) Non-geminate velar nasal ([N]) occurs only as the result of place assimilation.
(12) Evidence for knowledge of phonotactics
a. Perception of phonetically ambiguous stimuli (Pitt 1998, P&P, More- ton 2002, Cognition), word-spotting in continuous speech (Suomi, Mc- Queen & Cutler 1997, JML, McQueen 1998, JML, McQueen, Otake & Cutler 2001, JML, ) b. well-formedness ratings of novel words (Pierrehumbert 1994, LabPhon III, Coleman & Pierrehumbert 1997, ACL, Frisch, Large & Pisoni 2000, JML; Frisch & Zawaydeh 2001, Language) c. memory for novel words (Frisch, Large & Pisoni 2000, JML), naming latency of existing and novel words (Vitevitch 2002, JEP) d. See also studies on experimental phonotactics ( transitional proba- bilities): Dell, Reed, Meyer & Adams (2000, JEP), Goldrick (2002), others.
(13) Some relevant work
a. Albright & Hayes (2002): comparison of rule- and exemplar- based models. b. Wilson (2006): testing for biases in acquisition of experimental alter- nations. (See also Finley 2007)
(14) Brief review of English past tense
a. Regular (weak) verbs: /-d/, /-t/, /-@d/ (conditioned by the final con- sonant of the stem) b. Irregular (strong) verbs: Various degrees of irregularity; irregulars with same change often bear ‘family resemblance’ to one another. (see Pinker & Prince 1989, p.185)
the change is made. ∅ → d / [SaIn ][+past] ∅ → d / [kn "
faIn ][+past] ∅ → d / [X[-son,+cont,-voice]aIn ][+past] c. General rules do not replace specific rules (all are available to form past-tenses).
(18) Novel findings of A&H
a. Well-known that irregular verbs fall into family-resemblance classes (Bybee & Slobin 1982, Language, Pinker & Prince 1988, Connections and Symbols). Ex. Many verbs that end in [IN] form their past tense with the I→[æ]+past change. b. A&Hs learner discovered the same type of structure in the set of regular verbs. Ex. All verbs that end in voiceless fricatives ([f,T,s,S]) are regular. c. A&Hs experimental results (generation of past-tenses and ratings) sug- gest that speakers are sensitive to such sub-regularities for both regular and irregular changes.
(19) Comparison with the GCM
a. Overall, the A&H theory fits the data somewhat better than the GCM. b. The A&H theory places an (innate) restriction on the set of rules that can be learned. (i) Only context local to the change can be used to generalize a rule. (ii) In other words, internal variables are not allowed (e.g. ∅ → d / [pXn ] is impossible). c. But the GCM computes the global similarity of a verb to all existing verbs. (i) The model is not biased to use contextual information local to the change. (ii) Therefore, it draws bizarre analogies (e.g., *[rEnd@r@d] based on similarity of ‘render’ to rend, end, vend, raid, fend, mend, tend, round, dread).
(20) Conclusions
a. Correct (morpho-) phonological generalization depends on structured similarity. Concepts such as locality, natural classes, etc. provide the structure for generalization. Many of these concepts were fundamental concerns of early generative phonology. b. Several dimensions on which traditional psychological and phonologi- cal theories differ. Most interesting proposals (such as A&Hs theory) set universal restrictions on generalization within a framework that is
more inductivist than traditional generative phonology.
(21) The substance question What role do substantive relations among phonological entities—relations such as similarity, enhancement/antagonism, and ease of transition—play in phonological systems?
(22) Synchronic grounding Substantive relations place restrictions and/or biases on the generalizations that are acquired by language learners. See SPE (ch. 9), Natural Phonology (Stampe 1979), phonetically-driven phonology (e.g., Steriade 2001,2002), Hayes Kirchner, and Steriade (eds.) (2004) and Wilson (2006) for related proposals.
(23) Diachronic grounding Substantive relations play a role in determining likely phonological changes, but do not restrict or bias synchonic generalizations. See Anderson (1981, LI), Hyman (1998), Blevins & Garrett (1998, Language), Buckley (2000), Blevins (2004) for related proposals.
(24) General conclusions
a. Determining how innate restrictions, or analytic biases, structure the generalizations that are induced from data is arguably the central prob- lem facing cognitive science. b. Phonological theory provides a framework of hypothesized representa- tions and analytic biases, both closely related to phonetic substance, within which to address this problem (e.g. Wilson’s (2006) substantive bias hypothesis).
1.4 Overview of the next few weeks
(25) Writing phonological rules, notation, principles of analysis, etc.
(26) Dimensions of complexity
a. Unnaturalness No synchronic substantive relation between a property of α and the context in which it is conditioned “random rules” b. Non-locality The conditioning element is not strictly local to α. The distance be- tween the two may be unbounded, possibly with restrictions on inter- veners.
(1) Feature Theory is the theory whose subject is subsegmental structure. Major issues include: a. How are features defined? b. What are the relevant features? c. What values can features take? d. Are there dependency relations between features and if so what are they? e. What is the role of feature underspecification? f. Where do features come from? g. How do we find answers to these questions?
(2) Goals of Distinctive Feature Theory:
a. Determine which units can be minimally contrastive in the world’s lan- guages (i) The idea is each language draws its distinctive features from this universal set b. Another idea is that the features bridge the gap between the abstract phonemic symbol and the physical realization of the sound—they are concrete instructions to articulators, for example.
(3) Of course there is always a tension:
a. The more features posited, the more phonetic detail is accounted for, and potentially more phonological processes can be explained with ap- peal to phonetic naturalness b. However, smaller inventories are simpler and make describing phono- logical processes easier
(4) The methodology:
a. If two sounds contrast minimally in some way in some language, then a feature must be posited to account for the difference. b. If a phonological process applies to some class of sounds then those sounds c. The more common the distinction is used across languages, the more languages which refer to the same class of sounds, is evidence for that feature and/or natural class. d. Consequently, the phonemic inventories and phonological processes found in the world’s languages constitute the major sources of evidence for distinctive feature theory. e. Increasingly detailed phonetic work providing articulatory (via ultra- sound or other more-invasive means) or acoustic measures can help clarify which speech sounds meet which definitions.
(5) The central idea behind features is that each segment is actually definable in terms of a set of properties. These properties are the features and this set is often called a feature bundle.
(6) Thus words, which we often think of as a string of segments, can actually be represented as a string of feature bundles. Here is one set of features we might use to illustrate this idea with the word mop.
[mop] =
−syllabic +sonorant +stop +nasal +LABIAL +voice
+syllabic +sonorant −stop −nasal +low +back −round
−syllabic −sonorant +stop −nasal +LABIAL −voice
(7) The features can be thought of as members of a set of elementary categories which combine to form the speech sounds of language.
(8) Where do the features come from?
a. It is standardly assumed that they are universal and innate (which