Introduction Regular system: for every input, the grammar produces only one output Ways to achieve...
-
date post
21-Dec-2015 -
Category
Documents
-
view
214 -
download
0
Transcript of Introduction Regular system: for every input, the grammar produces only one output Ways to achieve...
Introduction
Regular system: for every input, the grammar produces only one output
Ways to achieve regularity•Minimize competition between generalizations
Restricting structural description of rules so that no rules compete for the same input
•Resolve competition in a winner-take-all mannerStrict ranking in Optimality Theory
Linguistic description: Maximizing regularityThis paper: Some cases where speakers do not
Case I: Velar palatalization in Russian (Kapatsinski 2010a)
Rule: {k;g}{t ; }/_-itᶘ ᶚ j
Native lexicon: No exceptions
Nonce borrowings (web data):
Palatalization fails often
(e.g., to book bukitj)
Why?
Hypothesis: Russian speakers acquire an “overly”-general rule “just add –itj” that competes with {k;g}{t ; }itᶘ ᶚ j
The more –itj attaches to non-velars, the more reliable “just add –itj” will be.Thus the more likely it will be to outcompete {k;g}{t ; }itᶘ ᶚ j
if competition is resolved stochastically.
Prediction: palatalization should fail in front of suffixes that often attach to stems that end in consonants that are ineligible for palatalization
Crucial test case: Masculine diminutives
Three suffixes: -ik, -ok, -ek.
In the native lexicon, palatalization always applies to velars
Prediction confirmed
ConclusionsSystems that look regular from a linguistic description sometimes aren’t for the language users.
Human language learners do not maximize regularity to the extent that linguists do.
Thus the learned grammars do not provide as much information about the correct output as the grammar a linguist would generate.
To cope with ambiguity in the output of the grammar, learners rely on lexical retrieval whenever they can.
Future work: When are seemingly regular systems not? (e.g., Bybee 2008: morphologization reliance on retrieval)
Vsevolod KapatsinskiDepartment of Linguistics, University of Oregon
ReferencesAlbright, Adam, & Bruce Hayes . 2003. Rules vs. analogy in English past tenses:
A computational/experimental study. Cognition, 90, 119–161.
Butterworth, B. 1983. Lexical representation. In B. Butterworth (ed.), Language production ( Vol. II ): Development, writing, and other language processes, 257–294. London: Academic Press.
Bybee, Joan. 2008. Formal universals as emergent phenomena: The origins of structure preservation. In Jeff Good (ed.), Linguistic universals and language change, 108–121. Oxford: Oxford University Press.
Halle, Morris. 1973. Prolegomena to a theory of word formation. Linguistic Inquiry, 4, 3-16.
Kapatsinski, V. 2010a. Velar palatalization in Russian and artificial grammar: Constraints on models of morphophonology. Laboratory Phonology, 1, 361-393.
Kapatsinski, V. 2010b. What is it I am writing? Lexical frequency effects in spelling Russian prefixes: Uncertainty and competition in an apparently regular system. Corpus Linguistics and Linguistic Theory, 6, 157-215.
Zuraw, Kie. 2000. Patterned exceptions in phonology. Los Angeles, CA: UCLA dissertation.
Regularity is overrated: Stochastic competition in grammar and the primacy of the lexicon
Mean 0%Mean 1% Mean 35%
-ik mostly attaches to non-velars
-ik is the only diminutivesuffix in frontof which velarpalatalizationoften (35%) fails
Velar palatalization in Pseudo-Russian: Replicating the effect in the lab (Kapatsinski 2010a)
Native English speakers
Exposed to one of the following languages:
Same examples of palatalization
More support for non-palatalizing rule in LgII
Should see less velar palatalization in LgII ifthe rule is overgeneral(applies to velars) andif the competition withthe palatalizing ruleis resolved stochasticallysince the palatalizing ruleis stronger in both languages
*
Result
Case II: A regular spelling rule in Russian (Kapatsinski 2010b)
Why look at orthography?
Orthography is designed to be and taught as a regular rule system
The rules:
In Russian, obstruents devoice before voiceless obstruents
For prefixes ending in /z/, the devoicing is supposed to be reflected in the spelling
Prefixes ending in other consonants and all stems, including prepositions, always have a constant spelling independent of phonological context
Summary of Case IA linguistic description of the morphophonology of velar palatalization has rules that are too specific compared to what the learners acquire
Competition between rules is resolved by the learners stochastically (they don’t always go for the most reliable rule)
For native Russian speakers, velar palatalization is not very productive before –i and –ik. How come the lexicon contains no exceptions to the rule then?
I suggest the speakers rely on lexical retrieval to produce the forms that seem to be obeying the rule (Butterworth 1983, Halle 1973, Zuraw 2000, Albright & Hayes 2003). Plus, rule-violating forms may be perceived as awkward (Zuraw 2000).
The relevant grammar
Since the preposition and the prefix are spelled differently, this may make bez- especially hard to spell if the writer is uncertain whether they are spelling a PP or an adjective at some processing stage, compared, e.g., to the verbal prefix raz-, which does not have a corresponding preposition it could be confused with
This would in turn cause Russian writers to rely on retrieving orthographic forms of adjectives from memory
A stem should have a constant spelling
Don’t spell ‘z’
A phoneme should havea constant spelling
Spell what you hear
The preposition bez (always spelled with ‘z’)
The adjectives seem derived from the PP’s: A bez- adjective always has a corresponding PP but sometimes lacks a corresponding bez-less adjective
Google databez- raz-
Much higher error rate for bez-Much stronger correlation of error rate with word frequency for bez-, esp. within lexemes:
e.g., countless .PL is more common than countless.SG and is spelled more accurately too
Bez- Raz-Bez-
Conclusion: Reliance on inflected form retrieval to spell bez- but not raz-
Experimental dataA classroom dictation task. Graded. Non-speeded.Subjects should retrieve the rule if they know it.Subjects are college students, extensively trained on the rule.No errors for raz- but at chance for bez- with unknown words:
Summary of Case IIWhile the same rule describes the spelling of bez- and raz-, the spelling of bez- relies on lexical retrieval rather than rule application but the spelling of raz- is largely rule-based
This is likely due to competition between the spelling rules or orthographic forms of bez-the-prefix and bez-the-preposition, which are probably the same lexical entry
There may be competition even in what looks like a regular system when the applicability of the rule is difficult to evaluate in processing due to uncertainty regarding whether the input meets the structural description
Adjectival inflection is regular whole wordforms need not be stored
Masculine singular forms predictable given stress