MPhil Seminar: Evaluating OT
Two attacks on constraints
OTROTB-LO: no constraints on URs
Reiss, NoBanana: no surface (or other) constraints
Basic problem:
 Can one extract generalizations from surface (especially static/non-alternating)
Evidence for extraction of generalizations from the lexicon
Best-known case: goed stage of L1 acquisition
Also Ohala, Pierrehumbert, Hayes, etc. on statistical knowledge
Marcus et al. 1999, Guasti 2002, Kuhl 2004 on child language: infants are able
to perform statistical analysis over pre-lexical representations, e.g. compute
distributional regularities and find the most frequent word shapes.
Humans form phonological generalizations over their lexicons, often best
modelled as MSCs or surface constraints
 Often statistical in origin, but may be deterministic
OTROTB-LO wrongly predicts this to be impossible and creates other problems
Where are linguistic
generalizations captured?
lexicon/underlying representation
Hale and Reiss: only
here (no constraints)
(GEN; no generalizations)
¿constraints on GEN?
surface representation
Morpheme Structure Constraints
Initially employed to capture static phonological
generalizations about morpheme structure, as
opposed to alternations being captured by rules
Root Harmony (Kiparsky 1968)
 C0 V[atr] … C0 V[atr] C0 (Akan and Wolof, K 1994:351)
Japanese: all post-nasal obstruents must be voiced in
native words
 tombo ‘dragonfly’ (*tompo)
 mi-te ‘seeing’ vs. šin-de ‘dying’
 Can be modeled as an OT output constraint *NT
(though Itō, Mester, and Padgett 1995:819 call it an MSC…)
See Kenstowicz and Kisseberth 1979:425-433, Kenstowicz 1994:351-3, 524-8 for discussion
Early arguments for MSCs
Halle 1959, 1962, Chomsky and Halle
1968, etc.
account for native speakers’ intuitions of what
constitutes a well-formed word in their
Esper 1925
Ss learn names of 16 objects, each having one of four different shapes and one of four different colors
Ss trained on 14 object-name associations but tested on 16 to see if they generalize what they learned
3 experimental conditions:
names presented to Group 1:
Names presented to Group 2:
naslig, sownlig, nasdeg, sowndeg, where nas- and sown- coded color and -lig and -deg coded shape
Since these names consisted of two phonologically legal morphemes, this group could simplify their task by learning not 16
names but 8 morphemes (if they could discover them) plus the simple rule that the color morpheme preceded the shape
morpheme in each name.
bi-morphemic names, as with Group 1
unlike group 1, the morphemes were not phonologically legal for English, e.g., nulgen, nuzgub, pelgen, pezgub (where nu- and
pe- were color morphemes and -lgen and -zgub were shape morphemes, the latter two violating English morpheme structure
Names presented to Group 3 (a control group):
names with no morphemic structure
no recourse but to learn 16 idiosyncratic names
As expected, group 1 learned their names much faster and more accurately than group 3.
Performance of Group 2 was similar to (and marginally worse than) that of group 3
Analysis of the errors of group 2, including how they generalized what they’d learned to the two object-name associations
excluded from the training session, revealed that they tried to make phonologically legal morphemes from the ill-formed ones.
Demonstrates (i) psychological reality of MSCs; (ii) ability to conduct morphological analysis
languages do not always make URs conform to surface phonotactics, e.g. Homshetsma ‘hit’, Maori final consonants, Hebrew
consonantal roots, Turkish epenthesis
the semantic/ morphological shape of the compound words to be learned in this system is basically unnatural. It is unnatural
because languages almost never attach color words to shape words to form compounds or derived words (Tahny, 1977).
Although we occasionally find "frozen forms" (Newport & Bellugi, 1978) like greenhouse or blackbird, we almost never find a
productive process that turns the concept "red square" into the single word "redsquare."
Arguments against MSCs
Duplication Problem (Kisseberth 1970 et seqq.)
(i)  Turkish
(Kaun and
Harrison 1999)
(ii)  Marash
(Vaux 1998)
Japanese MSC *NT for *tompo and rule for šin-de vs. mi-te
“there is good reason to doubt the basic assumption…that the harmony found in
roots and affixes is the product of two separate grammatical mechanisms: a
morpheme structure condition and a feature-changing rule…it implies the existence
of [i] languages in which all the suffixes systematically harmonize to the root but the
roots show no restrictions on vowel combinations or in which the opposite state of
affairs holds (i.e. [ii] the root vowels harmonize but affixes fail to alternate).” (K
“this formal similarity and functional redundancy between MSCs and rules is a
significant liability of the classic theory. If MSCs and rules really are distinct
components of linguistic theory, then they should be cleanly differentiated in form
and function, but they are not.” (McCarthy 1998)
“This stance makes maximal use of theoretical resources already required, avoiding
the loss of generalization entailed by adding further language-particular apparatus
devoted to input selection. (In this we pursue ideas implicit in Stampe 1969,
1973/79, and deal with Kisseberth’s grammar/lexicon “duplication problem” by
having no duplication.)” (P and S 1993/2002:209)
Wellformedness judgements
MSCs predict that speakers can only make ternary distinctions in well-formedness,
whereas speakers in fact make scalar judgements (Greenberg and Jenkins 1964,
Ohala & Ohala 1986:242; see Pierrehumbert 2003 for literature review).
Faulty conception of MSCs I
Kie Zuraw presents typical OT misconception that MSCs are required to capture any
surface-true generalization
Zuraw’s take on DP
analysis of these data:
• MSC for ‘green’ etc.
• rule for ‘old man’ etc.
She sees this as “Duplication
Problem” (!)
Actual DP analysis
(assuming new loans are
• Single rule for both ‘green’
and ‘old man’ etc.
• Not subject to DEC
Faulty conception of MSCs II
McCarthy 1998
“According to the premises of classic generative phonology, final devoicing in L is
a result of a phonological rule. In L’, though, devoicing is attributed to a
morpheme structure constraint (MSC), the name given to restrictions on
underlying representations.”
BV: in the absence of evidence from loanwords, language games, etc. showing
that the lack of final D in URs is the product of an active MSC (which McCarthy
doesn’t provide), such cases actually involve “Stampean Occultation”:
 “Suppose some rule consistently replaces the structure /A/ by [B]. Finding no surface
• A = voiced stop
• B = voiceless stop
[A]s, language learners will not be tempted to set up underlying /A/s in the lexicon,
positing only underlying /B/s instead. In this way, /B/ hides or ‘occults’ /A/, obtaining
the same descriptive effect as an anti-/A/ MSC without invoking any actual restrictions
on the lexicon.” [McCarthy 1998:1]
Faulty conception of MSCs III
“Under the thesis of richness of the base, OT does not
countenance morpheme structure constraints. This paper
shows that some phenomena that have been attributed
to morpheme structure constraints can be analyzed with
constraints that forbid alternations within paradigms.”
Given what I’ve already proposed,
how do you think we should deal with
Dialect B?
ROTB and Lexicon Optimization
“OT attributes linguistic generalizations to the grammar, not the
lexicon...this thesis is called ‘richness of the base’: inputs are
unrestricted, but the grammar is responsible for mapping all inputs
onto pronounceable forms of the language.” (McCarthy 2003:53)
“if the grammar yields an inventory with only unvoiced obstruents, no
segments in lexical forms will contain [voice] without [sonorant] — even
though all feature combinations are universally available as inputs.”
(Smolensky 1996)
Lexicon Optimization (Inkelas 1994, based on P&S 1993/2002:209)
“Given a grammar G and a set S = {S1, S2, ... Si} of surface phonetic forms
for a morpheme M, suppose that there is a set of inputs I = {I1, I2, ... Ij},
each of whose members has a set of surface realizations equivalent to S.
There is some Ii  I such that the mapping between Ii and the members of
S is the most harmonic with respect to G, i.e. incurs the fewest marks for
the highest ranked constraints. The learner should choose Ii as the
underlying representation for M.” (Inkelas 1994)
ROTB and Lexicon Optimization
Turkish final devoicing (to be discussed in more detail later)
[vɑth] ‘watt’ : [vɑthɯ] ‘watt-accusative’
[thɑth] ‘taste’ : [thɑdɯ] ‘taste-accusative’
One can force UR  SR by having alternations in the paradigm (P&S
1993/2002:210, Inkelas 1994:7), but if there is no evidence for alternations
(e.g. with a nonce word), ROTB-LO (wrongly) predicts UR = SR.
Voice  Coda
[vɑth] : [vɑthɯ]
[vɑth] : [vɑdɯ]
[vɑd] : [vɑdɯ]
Voice  Coda
[thɑth] : [thɑthɯ]
[thɑth] : [thɑdɯ]
[thɑd] : [thɑdɯ]
LO cases
Voice  Coda
Voice  Coda
 /thyb/
 /gruph/
Response to response to MSCs
The duplication argument, which is the heart of the
attack on MSCs, only holds ceteris paribus, but in fact all
else is not equal
ROTB-LO incorrectly predicts the nonexistence of productive
lexical generalizations utilized by speakers in constructing
underlying representations.
ROTB-LO incorrectly predicts (assuming universal markedness
constraint hierarchies; cf. Prince and Smolensky 1993, Steriade
1999:42, Lombardi 2003) the absence of languages containing
the marked but not the unmarked member of a phonemic
 Cf. Russian has palatalized /čj/ but not plain */č/
ROTB-LO incorrectly predicts conformity of URs to surface
ROTB-LO incorrectly requires full spec. in non-alternating cases
ROTB-LO requires stipulation that certain GEN alterations (e.g.
syllabification) are invisible to Faith and Ident constraints
Predictions for picking UR from ambiguous input:
OTROTB-LO: pick transparent UR
Hayes 1995: pick base form as UR whether or not there are alternations
Gallistel 2003: When animals and humans have to solve problems with
incomplete knowledge, they use stochastic/probabilistic models
 NB deterministic generalization may be spawned from statistical knowledge
 In a language with 60% s and 40% t, s may be picked 60% or 100% of the time;
choice may be arbitrary with insufficiently skewed statistics, e.g. with pigeons
Type 1 (structure-preserving)
English final /r/
 Several nonrhotic Englishes productively assign final /r/ to all low-vowel-final roots
(Mohanan 1985, Stampe 1991, Harris 1994)
English backformation wrt Velar Softening (Pierrehumbert 2002)
 2 subjects backformed e.g. hovacity  hova[k], 33% and 75% of the time
Devoicing languages (German, Russian, Polish; Turkish, Lac Simon, Dutch)
Korean word-final [t]  /s/
Japanese [ŋ]  /g/ (Ito, McCarthy)
Type 2 (non-structure-preserving)
English flapping
 sporadic for some: antidote for anecdote, calisthentics, etc.
 systematic: SN’s flap  /t/
Korean borrowing of Coda [t]
Korean word-final [t|]  /t, th, t’, č, čh, č’, s, s’/
Surface word-final postvocalic [t] in loans and nonce words
invariably assigned to /s/ (Martin 1992, Kang 1998, Hayes
1998, Iverson & Lee 2004)
supermarket  nom. [supəmakhet|], dat. [supəmakhese]
OTROTB-LO wrongly predicts assignment to /t/
basic problem: OTROTB-LO does not allow for statistical
generalizing over the lexicon to play a role in the
construction of URs
What appears to be involved in the Korean case is that
speakers know that surface word-final [t]s most often
come from underlying /s/ in their native lexicon, and they
therefore assign new words to the same pattern.
NB these override voice
specification in source language
Turkish final [voice]
voiced UR hits
voiceless UR
E tube
E club
145,000 kulüpü
17 (0.1%)
E/F group(e) gurubu
F principe
327,000 grupu
448 (0.1%)
All [polysyllabic] forms that have a voiceless obstruent when
final have a voiced one when suffixed (Lewis 1967:11)
The converse has now developed for monosyllables (Inkelas,
Pycha, and Sprouse 2004)
TELL: 19 monosylls with final voiced stop; 145 with
voiceless; current MSC plausibly extracted from this
Lac Simon Algonquian
1. underlying voicing contrast
2. rule of initial obstruent devoicing
3. all new stem-initial obstruents  underlyingly voiced
(Nykiel and Nykiel 1979, Kaye 1979, Iverson 1983).
French banane [banan]  LSA [pa:na:n] ‘banana’, but
nba:na:nm ‘my banana’
English coffee  LSA [ko:fi:ke] ‘he makes coffee’, but
nigo:fi:ke ‘I make coffee’
SR w/ devoicing
SR w/o devoicing
ka:zo:tm ‘he hides’ n-ga:zo:tm ‘I hide’
n-ka:t ‘my leg’
NB the relevant frequency facts
for Lac Simon are not known.
not *n-ga:t; note that the same
1st person prefix conditions the
voiced allophone in (a)
Statistical knowledge
The basic problem:
OTROTB-LO does not allow for generalizations extracted from
statistical properties of the lexicon to play a role in the grammar
Counterevidence (cf. Skousen 1989):
 Greenberg and Jenkins 1964, Ohala and Ohala 1986, Frisch, Large,
and Pisoni 2000, Hay, Pierrehumbert, and Beckman 2004, etc. etc. on
the well-formedness of English nonce words
 Hayes 1995 on Turkish
 Pierrehumbert 2002 on English velar softening
 Polish speakers assign masculine gender to all consonant-final words
and feminine gender to all [a]-final words (Baran 2000)
Statistical knowledge  (categorical?) linguistic generalizations:
 “All other things being equal, the cognitive system prefers
generalizations which yield more information about the outcome over
those which yield less.” (Pierrehumbert 2002)
 “speakers extend morphological patterns based on abstract structural
properties, of a kind appropriately described with rules” (Albright and
Hayes 2003)
More deterministic…
German chooses -s as its productive plural, though it
isn’t most frequent (though frequency does affect its
productivity–Bybee 1995)
Moreton 1999:
English speakers aware of MSC banning final lax vowels
“phonotactic knowledge consists of categorical, rule-like
prohibitions, rather than emerging from statistical properties of
the lexicon”
Inkelas, Pycha, and Sprouse 2004 on Turkish voice
alternations: not conditioned by lexical neighborhood
density or frequency
mono- vs. polysyllabicity is best predictor of (non-)alternation
Underlying -structure
“CV-language learners will never insert into the
lexicon any underlying forms that violate the
(surface) syllable structure constraints of their
language” (P&S 2002:210)
Problem: Turkish and other languages that do
not postulate underlying epenthesis, even
though doing so does not conform to their
surface syllable canon
vakith ‘time’ : acc. vakth-i (< Arabic wakt)
istop ‘stop’ : acc. istop-u
 (not *istob-u, the expected polysyllabic treatment)
ROTB-LO requires full specification
in non-alternating cases
OTROTB-LO requires that all non-alternating
surface forms have fully specified lexical entries
Disproven by Kaun and Harrison 1999 with
respect to root-internal harmony in Tuvan,
Finnish, and Turkish
After application of relevant language games,
harmonic roots re-harmonize but disharmonic roots
Cf. Krämer 2004 for German glottal stop
insertion and English laxing
(He argues that LO actually can’t decide between fully
specified and underspecified form as UR, since
identity constraints are stipulated to not penalize
underspecified URs)
Incorporating ROTB into OT requires stipulating that GEN
be able to alter inputs in ways that are invisible to
faithfulness constraints (McCarthy 2002:38) and Ident
constraints (Krämer 2004).
McCarthy 2002:38: this is the only way to account for the universal
non-contrastiveness of certain phonological distinctions
 syllabification of tautomorphemic sequences is never contrastive, e.g. vs. ha.bla
 “A necessary condition for ensuring that syllabification is never
contrastive is that syllabification is faithfulness-free, so an unsyllabified
input like /maba/ or a syllabified input like /mab.a/ will be associated
by GEN with all of the following fully faithful and fully syllabified
candidates: m.a.b.a, ma.b.a,, m.aba, m.ab.a,, mab.a,
maba. Many of these candidates are sure losers for markedness
reasons, such as the absurd monosyllable maba. But they are still fully
faithful in the sense that they incur no faithfulness violations.”
ROTB doesn’t follow from
OT architecture
IO constraints allow reference to input forms
 OT has the power to evaluate I constraints
(constraints on inputs without reference to
corresponding outputs)
in fact, these are less computationally complex than
IO constraints
*{#[…D]#}I : no monosyllabic URs ending in
voiced obstruent
NB I constraints don’t do any work in IO
mappings; only involved in UR construction
 /thyph/
Voice  Coda
Summary of OTROTB-LO problems
1. Incorrectly predicts the nonexistence of MSCs
2. Incorrectly predicts the absence of languages
containing the marked but not the unmarked
member of a phonemic opposition
3. Incorrectly predicts conformity of URs to
surface phonotactics
4. Incorrectly requires full specification in nonalternating cases
5. Stipulates invisibility to Faith and Ident
The problems presented here are resolved
straightforwardly by assuming that humans can extract
generalizations from the structure of their lexicon.
NB generalizations can be extracted in the absence of
alternations (cf. Dell et al. 2000), e.g. from statistical knowledge
This move is consistent with what we know about
human and primate cognition:
Pierrehumbert 2002, 2003, etc. on statistical knowledge in
Marcus et al. 1999, Guasti 2002, etc. on child language
Kirkham et al. 2002 on vision in infants
Ramus et al. 2000 and Hauser et al. 2002 on primates
Grounded in the fundamental linguistic tenet that
extracting generalizations is the heart of grammar
Surface constraints
Dell et al. and Goldrick 2004 on speech
errors (as we saw in the speech errors
NB implies that humans can learn constraints
on representations in the absence of
alternations (cf. English learning of h and
engma distribution)
Identity constraints and
ineffability (Control constraints)
schm reduplication
Q19 Schmuck
Q20 Schmooze
Ø (70), shluck (8), schnuck (5), schmuck (4), fluck (3), shpuck
(1), fuck, smuck, shfuck, shvuck, schmluck, shnook
Ø (59), shnooze (10), shmooze (4), flooze (4), shpooze (4),
shlooze (3), shmmooze, commooze, shplooze, mooze, wooze
Q22 Schmidt
Ø (66), shlidt (4), shpidt (4), shmidt (3), shnidt (3), flidt (2),
vlidt, smidt, midt
morpheme sequencing
German Berlin-er ‘person from Berlin’ vs. Münster-aner
Lenin-akan-yan vs. *Lenin-akan-akan
Humans can and do extract constraints
(both surface and underlying) from
phonological and morphological data (both
alternating and static)
Important component of animal cognition: cf.
conditioning studies
NB at least some constraints are inviolable
Theories attacking such constraints
(especially OT) misunderstand use of MSCs
and ignore much of the relevant data.
References I
Booij, Geert. 1999. Morpheme structure constraints and the phonotactics of Dutch. In Harry van der Hulst & Nancy Ritter (eds.) The Syllable. Views and Facts.
Berlin/New York: Mouton de Gruyter, 53-68.
Christdas, P. 1986. Morpheme Structure Constraints and Underspecification. Paper presented at the annual meeting of the Linguistic Society of America, New
York, NY.
Dell, Gary, K. Reed, D. Adams, and A. Mejer. 2000. Speech errors, phonotactic constraints, and implicit learning: a study of the role of experience in language
production. Journal of Experimental Psychology (LMC) 26.6:1355-1367.
Dinnsen, Daniel and Laura McGarrity. 2004. On the nature of alternations in phonological acquisition. Studies in Phonetics, Phonology, & Morphology 11:23-42.
Esper, Erwin. 1925. A technique for the experimental investigation of associative interference in artificial linguistic material. Language Monographs, no. 1.
Goldrick, Matthew. 2004. Phonological features and phonotactic constraints in speech production. Journal of Memory & Language 51.4:586.
Greenberg and Jenkins 1964. they derive well-formedness judgements from comparison to forms in the lexicon, not MSCs
Halle, Morris. 1959. The Sound Pattern of Russian. The Hague: Mouton.
Hauser, Marc, Daniel Weiss, and Gary Marcus. 2002. Rule learning by cotton-top tamarins. Cognition 86:B15-22.
Hayes, Bruce. 1998. On the Richness of Paradigms, and the Insufficiency of Underlying Representations in Accounting for them. Lecture presented at Stanford
Inkelas, Sharon. 1994. The Consequences of Optimization for Underspecification. In NELS 25. 287-302.
Itô, Junko, and Armin Mester. 1999. On the Sources of Opacity in German. Coda Processes in German. Manuscript, UCSC and responded to in van Oostendorp
Iverson, Greg. 1983. Voice Alternations in Lac Simon Algonquin. Journal of Linguistics 19:161-164.
Iverson, Greg. 2004. Deriving the Derived Environment Constraint in Non-Derivational Phonology. Studies in Phonetics, Phonology and Morphology 11:1-23.
Katamba, Francis and Larry Hyman. 1991. Nasality and morpheme structure constraints in Luganda. Africanistische Arbeitspapiere 25. 175-211.
Kaun, Abigail. 1998. Input constraints in Tamil. Paper presented at CLS 34, April 17, 1998.
Kaun, Abigail and David Harrison. 1999. Pattern-Responsive underspecification. In Proceedings of the 30th Conference of the North Eastern Linguistics Society.
Kawasaki, H. 1982. An acoustical basis for universal constraints on sound sequences. Doctoral dissertation, University of California, Berkeley.
Kaye, Jonathan. 1979. On the alleged correlation of markedness and rule function. In D. Dinnsen (ed.) Current Approaches to Phonological Theory, pp. 272280, Indiana University Press, Bloomington.
Keating, Patricia. 1985. Linguistic and nonlinguistic effects on the perception of vowel duration. UCLA working papers in phonetics 60:20-39.
Kenstowicz, Michael, and Charles Kisseberth. 1979. Generative phonology. San Diego: Academic Press.
Kerkhoff, Annemarie and Elise de Bree Kuhl. 2005. Acquisition of Morphophonology in Children with Specific Language Impairment and Typically Developing
Children. Ms., Utrecht.
Kirkham, Natasha, Jonathan Slemmer, and Scott Johnson. 2002. Visual statistical learning in infancy: Evidence for a domain general learning mechanism.
Cognition 83.2.B35–42.
Krämer, Martin. 2004. Optimal underlying representations. NELS 35, University of Connecticut.
Larson, M. 1982. Referring to one’s lexicon for judgments on morpheme structure constraints: evidence from the issue of frequency. Manuscript, San Jose, CA,
referred to in Ohala and Ohala 1986.
Marcus, Gary, S. Vijayan, S. Bandirao, and P. Vishton. 1999. Rule learning by seven-month old infants. Science 283:77-80.
References II
McCarthy, John. 1998. Morpheme structure constraints and paradigm occultation. In M. Catherine Gruber, Derrick Higgins, Kenneth Olson and Tamra Wysocki
(eds.) CLS 32, Part 2: The Panels. Chicago, IL: Chicago Linguistic Society. 123-150.
McCarthy, John. 2003. Comparative markedness [short version]. Theoretical Linguistics (to appear).
McCarthy, John. 2003. Richness of the Base and the determination of underlying representations. Ms., University of Massachusetts, Amherst.
Mester, Armin. 1992. Morpheme structure constraints. Topic article in International Encyclopedia of Linguistics, by W. Bright, Oxford University Press, Vol. 3, pp.
Moreton, Elliott. 1999. Evidence for Phonotactic Grammar in Speech Perception. Proceedings of the 14th Annual International Congress of Phonetic Sciences,
San Francisco.
Napoli, Donna Jo and Jeff Wu. 2003. Morpheme structure constraints on two-handed signs in American Sign Language: Notions of symmetry. Sign Language &
Linguistics 6.2:123–205.
Nykiel, J. and B. Nykiel. 1979. Loan words and abstract phonotactic constraints. Canadian Journal of Linguistics 24:71-93.
Ohala, John, and Manjari Ohala. 1986. Testing hypotheses regarding the psychological manifestation of morpheme structure constraints. In: J. J. Ohala, J. J.
Jaeger (eds.), Experimental phonology, 239-52. San Diego: Academic Press.
Paradis, Carole & Jean Franois Prunet (1993). On the validity of morpheme structure constraints. In Carole Paradis & Darlene LaCharit (1993). 235-256.
Pierrehumbert, Janet. 2003. Probabilistic phonology: discrimation and robustness. In R. Bod, J. Hay and S. Jannedy (eds.), 2002. Probability Theory in
Linguistics. Cambridge: MIT Press.
Port, Robert. 1981. Linguistic timing factors in combination. Journal of the Acoustical Society of America 69:262-74.
Port, Robert and Penny Crawford. 1989. Incomplete neutralization and pragmatics in German. Journal of Phonetics 17:257-82.
Ramus, Frank, Marc Hauser, Cory Miller, Dylan Morris, and Jacques Mehler. 2000. Language discrimination by human newborns and by cotton-top tamarin
monkeys. Science 288:349-51.
Redford, Michael. 2000. The question of Inputs in OT: A constraint-based, computational model of English meter. Manuscript, Leiden University.
Reiss, Charles. 2000. Optimality Theory from a Cognitive Science Perspective. The Linguistic Review.
Saffran, Jenny, Richard Aslin, and Elissa Newport. 1996. Statistical learning by 8-month-old infants. Science 274:1926-28.
Scobbie, J., John Coleman, and S. Bird. 1996. Key Aspects of Declarative Phonology.
Skousen, Royal. 1989. Analogical modeling of language. Dordrecht: Kluwer.
Smolensky, Paul. 1996. The Initial State and Richness of the Base in Optimality Theory. Technical Report, JHU CogSci 96 4.
Smolensky, Paul, Lisa Davidson, and Peter Jusczyk. 2000. The Initial and Final State: Theoretical Implications and Empirical Explorations of Richness of the
Base. ROA.
Stanley, Richard. 1967. Redundancy rules in phonology. Language 43:393-436.
Steriade, Donca. 1997. Phonetics in Phonology: The Case of Laryngeal Neutralization. Manuscript, UCLA.
Vaysman, Olga. 2002. Against Richness of the Base: Evidence from Nganasan (NAPhC 2, 2002)
Vijayakrishnan, K. 1984. Morpheme Structure Constraints on Vowels and the Prosodic Foot in Tamil. CIEFL Occasional Papers 1.
Wilbur, R. B. (1982) The development of morpheme structure constraints in deaf children. The Volta Review 84, 7-16.
Yip, Moira. 1998. Lexicon optimization in languages without alternations. In Current Trends in Phonology, Jacques Durand, ed. Paris: Royaumont.
Zimmer, Karl. 1969. Psychological correlates of some Turkish morpheme structure conditions. Language 45:309-321.
Ambiguity and animal wug tests

Formal and empirical arguments for Morpheme Structure