Rules, Constraints, and Overlapping Violations: the case of Acoma accent loss Approaches to Phonological Opacity GLOW Workshop 2006 Joan Chen-Main

Slides:



Advertisements
Similar presentations
Optimality Theory (OT) Prepared and presented by: Abdullah Bosaad & Liú Chàng Spring 2011.
Advertisements

TO ONSET OR NOT TO ONSET: THAT IS THE QUESTION Rina Kreitman Emory University – According to the Sonority Sequencing Principle syllables.
Optimality Theory Presented by Ashour Abdulaziz, Eric Dodson, Jessica Hanson, and Teresa Li.
Optimizing Compilers for Modern Architectures Allen and Kennedy, Chapter 13 Compiling Array Assignments.
1 Interaction between phonology and syntax in Icelandic Arguments for a strongly parellel OT-analysis A Phonological Workshop University of Iceland May.
The Sound Patterns of Language: Phonology
1 Phonology → Phonetics Understanding Features 2 Richness of the Base The source of all systematic cross-linguistic variation is constraint reranking.
Optimality Theory Abdullah Khalid Bosaad 刘畅 Liú Chàng.
Lecture 4 The Syllable.
Optimality Theory (Prince & Smolensky 1993). Outline Phonetics and Phonology OT Characteristics Output-Oriented Conflicting Soft Well-formedness Constraints.
Tone, Accent and Stress February 14, 2014 Practicalities Production Exercise #2 is due at 5 pm today! For Monday after the break: Yoruba tone transcription.
Gestural overlap and self-organizing phonological contrasts Contrast in Phonology, University of Toronto May 3-5, 2002 Alexei Kochetov Haskins Laboratories/
Part Four PHONOLOGICAL PROCESSES.  Speech sounds are by nature dynamic and flexible, and highly susceptible to the influence of the ‘environment’, i.e.
A Study of Speech Perception: Julie Langevin Communication Sciences and Disorders Faculty Mentor: Timothy Bryant The Psychological Reality of the Obligatory.
Autosegmental Phonology
History of Phonology with an emphasis on recent history.
Distributional Cues to Word Boundaries: Context Is Important Sharon Goldwater Stanford University Tom Griffiths UC Berkeley Mark Johnson Microsoft Research/
Introduction Regular system: for every input, the grammar produces only one output Ways to achieve regularity Minimize competition between generalizations.
Research on teaching and learning pronunciation
January 24-25, 2003Workshop on Markedness and the Lexicon1 On the Priority of Markedness Paul Smolensky Cognitive Science Department Johns Hopkins University.
Explanation for Language Universals Marta i Aleksandra.
Belief Revision Lecture 1: AGM April 1, 2004 Gregory Wheeler
Lecture 1 Introduction: Linguistic Theory and Theories
Overview Part 2 – Circuit Optimization 2-4 Two-Level Optimization
[kmpjuteynl] [fownldi]
…not the study of telephones!
Present Experiment Introduction Coarticulatory Timing and Lexical Effects on Vowel Nasalization in English: an Aerodynamic Study Jason Bishop University.
Ch 9 & Ch 10 Slide 1 Ch 9 – Productivity Productivity – the capacity of a rule to apply to novel circumstances. P. 190 Vowel nasalization in English is.
Phonological Theory Beijing Foreign Studies University 2008.
Lecture 7 Syllable Weight. English Word Stress The account of English stress presented so far only works for a subset of nouns/suffixed adjectives and.
1 Speech Perception 3/30/00. 2 Speech Perception How do we perceive speech? –Multifaceted process –Not fully understood –Models & theories attempt to.
Copyright 2007, Toshiba Corporation. How (not) to Select Your Voice Corpus: Random Selection vs. Phonologically Balanced Tanya Lambert, Norbert Braunschweiler,
Phonological Theory.
Exam Taking Kinds of Tests and Test Taking Strategies.
Classification and Ranking Approaches to Discriminative Language Modeling for ASR Erinç Dikici, Murat Semerci, Murat Saraçlar, Ethem Alpaydın 報告者:郝柏翰 2013/01/28.
Ch 7 Slide 1  Rule ordering – when there are multiple rules in the data, we have to decide if these rules interact with each other and how to order those.
Linguistic Theory Lecture 10 Grammaticality. How do grammars determine what is grammatical? 1 st idea (traditional – 1970): 1 st idea (traditional – 1970):
Learning Automata and Grammars Peter Černo.  The problem of learning or inferring automata and grammars has been studied for decades and has connections.
Focus marking in monolingual and heritage Spanish: Preliminary results UIC Bilingualism Forum April 30, 2009.
Harmonic Ascent  Getting better all the time Timestamp: Jul 25, 2005.
Ch 3 Slide 1 Is there a connection between phonemes and speakers’ perception of phonetic differences? (audibility of fine distinctions) Due to phonology,
Handout #10 Alternations with ø. Kinyarwanda (Rwanda) ja˘ndika8 “he/she writes” iBitaBo8 “a book” ja˘ndik iBitaBo8 “he/she writes a book” umu˘nhu8 “a.
Bernd Möbius CoE MMCI Saarland University Lecture 7 8 Dec 2010 Unit Selection Synthesis B Möbius Unit selection synthesis Text-to-Speech Synthesis.
Evaluating prosody prediction in synthesis with respect to Modern Greek prenuclear accents Elisabeth Chorianopoulou MSc in Speech and Language Processing.
Assessment of Phonology
Handout #14 Lamba. Lamba (a Bantu language of Zambia) (Kenstowicz and Kisseberth 1979: 71-72)
Models of Linguistic Choice Christopher Manning. 2 Explaining more: How do people choose to express things? What people do say has two parts: Contingent.
Lecture 2 Phonology Sounds: Basic Principles. Definition Phonology is the component of linguistic knowledge concerned with rules, representations, and.
Hello, Everyone! Part I Review Review questions 1.In what ways can English consonants be classified? 2. In what ways can English vowels be classified?
The phonology of Hakka zero- initials Raung-fu Chung Southern Taiwan University 2011, 05, 29, Cheng Da.
Program Structure  OT Constructs formal grammars directly from markedness principles Strongly universalist: inherent typology  OT allows completely formal.
Ch 8 Slide 1 Some hints about analysis First try to establish morphemes. If there is allomorphy, list all of the alternants (remember some morphemes don’t.
Distributed Models for Decision Support Jose Cuena & Sascha Ossowski Pesented by: Gal Moshitch & Rica Gonen.
THE SOUND PATTERNS OF LANGUAGE
Principles Rules or Constraints
Intro to NLP - J. Eisner1 Phonology [These slides are missing most examples and discussion from class …]
Bridging the gap between L2 speech perception research and phonological theory Paola Escudero & Paul Boersma (March 2002) Presented by Paola Escudero.
Optimality Theory. Linguistic theory in the 1990s... and beyond!
1 LING 696B: Maximum-Entropy and Random Fields. 2 Review: two worlds Statistical model and OT seem to ask different questions about learning UG: what.
Compiler Construction Lecture Five: Parsing - Part Two CSC 2103: Compiler Construction Lecture Five: Parsing - Part Two Joyce Nakatumba-Nabende 1.
11 How we organize the sounds of speech 12 How we use tone of voice 2009 년 1 학기 담당교수 : 홍우평 언어커뮤니케이션의 기 초.
English Plurals FAITH (voi): Voicing must be same in input and output FAITH (voi): Voicing must be same in input and output FAITHV:Vowels in input and.
Lecture 4 The Syllable.
Lecture 7 Syllable Weight.
Step 1: Memorize IPA - practice quiz today - real quiz on Tuesday (over consonants)! Phonology is about looking for patterns and arguing your assessment.
G. Pullaiah College of Engineering and Technology
Basic Parsing with Context Free Grammars Chapter 13
Ordering of Hypothesis Space
Review.
Implementation of Learning Systems
Presentation transcript:

Rules, Constraints, and Overlapping Violations: the case of Acoma accent loss Approaches to Phonological Opacity GLOW Workshop 2006 Joan Chen-Main Johns Hopkins University

2 Rule-based Phonology Generalizations about a language’s sound structure stated as rewrite rules Opacity effects: generalizations are not surface true The application of the rules is ordered That’s all about the interaction of different rules, but...

3 Overlapping Violations What about when a single rule can apply multiple times to the same string? e.g. X → Y/__ X i.e. XX → YX independent: X X A X X → Y X A Y X overlapping: X X X A B → ?

4 Possible Ways to Apply a Rule Anderson’s (1974) How to Apply a Rule to a Form left-to-right –Mandarin tone sandhi3-3 → → right-to-left –Slovak ‘rhythmic law’ CV: CV: → CV: CV CV: CV: CV: → CV: CV CV simultaneous

5 Optimality Theory: the appeal Duplication in Rule-based theory –Rules are bound by the segmental inventory Conspiracies of Rules –Different repairs all resulting in the same surface form Do X except/only when...

6 Optimality Theory: difficulty 1 Opacity Opaque generalizations are non-surface true –take place at an intermediate representation –predicted by any theory allowing intermediate representations OT does not allow intermediate representations

7 Optimality Theory: difficulty 2 Non-local interaction Wilson’s (in prep) observation: In classic OT, unattested, non-local interaction is possible. e.g. between epenthesis and nasal spreading –*CC# –Spread-R([+nasal], PrWd): For every [+nasal] autosegment n, assign 1 violation for every segment in the same prosodic word that is to the right of n’s domain. (Walker,1998/2002, for Malay)

8 Optimality Theory: difficulty 2 Non-local interaction –Vowel epenthesis applies to a final consonant cluster except when there is a preceding [+nasal] feature anywhere in the word that is blocked from spreading to the right edge. /nawakast/ Spread-R([+nasal], PrWd) *CC# a. na  w  a  kas  t *****! b. na  w  a  kast *****

9 An OT Extension Targeted Constraint OT (Wilson, in prep) –does allow intermediate representations: solves opacity problem –avoids predicting “farsighted” patterns

10 Outline Targeted Constraint Optimality Theory (Wilson’s post 2001 version) –targeted constraints –candidate evaluation TCOT vs. rule-based phonology Acoma accent loss –difficulties with a rule-based account –availability of a TCOT account

11 Targeted Constraint OT Two novel aspects: 1)How change is integrated Changes are introduced by G EN s associated with a particular targeted markedness constraints. 2)How changes are evaluated The system rewards certain changes but penalizes others.

12 Targeted Constraints A targeted constraint C is a pairing of a locus of violation (λ) with a change (δ) Wilson’s claim: δ is the minimal perceptual change Machinery allows any rule to have a targeted constraint analogue

13 Targeted Constraint Analogues V [+short] → [-accent] / [+obst] __ [+obst] C 0 [+ syll, + accent] T:*C LASH : λ: two consecutive [+accent] syllables, where the first vowel is short and is flanked on both sides by an obstruent. δ: [+accent] → [-accent] in the first syllable C 0 = zero or more consonants

14 G EN C associated with a Targeted Constraint C G EN C maps a candidate to a candidate set: all candidates that can be derived by applying change δ to zero or more instances of the locus λ. e.g. If σ́σ́σ́ has 2 violations of T:*C LASH, then G EN T:*C LASH produces 4 candidates –the completely faithful candidate σ́σ́σ́ –the candidate where δ is applied to both λ’s σσσ́ –two candidates where δ is only applied to one λ σσ́σ́, σ́σσ́

15 Targeted Constraint Evaluation When comparing the input representation x with a candidate output y... –For every λ in y, assign 1 mark to y. –For every λ in x that is repaired in y in the way specified by δ, remove 1 mark from y. –For every λ in x that is repaired in y in a way not specified by δ, add 1 mark to y.

16 Example Evaluation candidate yviolations that remain violations fixed as specified by δ violations fixed, but not as specified by δ Total σ́σ́σ́ = +2 σσσ́ 0(-1) + (-1) 0 + 0= -2 σσ́σ́+1(-1) = 0 σ́σσ́ (-1)+1 + 0= 0 Input x: σ́σ́σ́

17 Rewarding Rule Application output yNumber of times rewrite rule could apply Number of times rewrite rule did apply (-2) for each rule application Total σ́σ́σ́ = +2 σσσ́ (-2)= -2 σσ́σ́ (-2)= 0 σ́σσ́ (-2)= 0 Input x: σ́σ́σ́

18 TCOT and Rule-based Phonology Specify preferred repairs Generate intermediate representations Good at accounting for data that is difficult for Classic OT How do TCOT and rule-based phonology differ?

19 TCOT vs. Rule-based Phonology Notion of minimal perceptual change (δ) –could be incorporated into rule-based theory Candidate evaluation in TCOT, as in Classic OT –competing output candidates evaluated against a set of ranked, violable constraints Does retaining an OT architecture ever result in different predictions?

20 Acoma: the people Acoma pueblo: Sky City –approx. 60mi west of Albuquerque, NM –oldest continuously inhabited city in the U.S. –distinctive pottery Acoma language –closely related to other pueblo languages –data from Miller’s (1965) book Dorothy Torivio

21 Acoma: accent ablaut (Miller 1965) In the presence of an ablauting suffix, a high accent is assigned to every syllable –sometimes with the lengthening of final vowel –sometimes excepting certain final syllables without accent ablaut: r û u n i š i ‘Monday’ with accent ablaut: r u u n i š i i z e‘on Monday’

22 Acoma: accent loss (Miller 1965) a short syllable between obstruents followed by an accented syllable loses its accent Anderson’s rule (1974) V → [-accent] / [+obst] ___ [+obst] C 0 [+ syll, + accent] –context: the conditions for application of the rule –focus: the segments that satisfy the conditions

23 One Focus for Accent Loss s i u k a c̉ a n i → s i u k a c̉ a n i‘when I saw him’ simultaneous left-to-rightright-to-left 1 focusok

24 Two Foci for Accent Loss k' a p i š ə́ n í → k' a p i š ə́ n i‘at night’ simultaneous rule application: σ́ σ́ σ́ σ́ rule applies to all foci σ σ σ́ σ́ attested pattern

25 Two Foci for Accent Loss k' a p i š ə́ n í → k' a p i š ə́ n i‘at night’ L  R rule application: σ́ σ́ σ́ σ́ rule applies to the leftmost focus σ σ́ σ́ σ́ rule applies to next-leftmost focus σ σ σ́ σ́ attested pattern

26 Two Foci for Accent Loss k' a p i š ə́ n í → k' a p i š ə́ n i‘at night’ R  L rule application: σ́ σ́ σ́ σ́ rule applies to the rightmost focus σ́ σ σ́ σ́ context for next-rightmost focus is destroyed σ́ σ σ́ σ́ non-attested surface pattern

27 Two Foci for Accent Loss simultaneous left-to-rightright-to-left 1 focusok 2 fociok X

28 Three Foci for Accent Loss s u c̉ i t i s t a a n i → s u c̉ i t i s t a a n i ‘when I was thinking’ simultaneous rule application: σ́ σ́ σ́ σ́ σ́ 3 foci for accent loss σ σ σ σ́ σ́ non-attested surface pattern

29 Three Foci for Accent Loss s u c̉ i t i s t a a n i → s u c̉ i t i s t a a n i ‘when I was thinking’ L  R rule application: σ́ σ́ σ́ σ́ σ́ rule applies to the leftmost focus σ σ́ σ́ σ́ σ́ rule applies to the next-leftmost focus σ σ σ́ σ́ σ́ rule applies to last focus σ σ σ σ́ σ́ non-attested surface form

30 Three Foci for Accent Loss s u c̉ i t i s t a a n i → s u c̉ i t i s t a a n i ‘when I was thinking’ R  L rule application: σ́ σ́ σ́ σ́ σ́ rule applies to the rightmost focus σ́ σ́ σ σ́ σ́ context for next-rightmost focus is destroyed σ́ σ́ σ σ́ σ́ rule applies to next-rightmost focus σ σ́ σ σ́ σ́ attested pattern

31 Inconsistent Rule Application simultaneous left-to-rightright-to-left 1 focusok 2 fociok X 3 fociXXok

32 Modified Rule Application Anderson’s solution –If any contexts for a rule contains a focus for the same rule, eliminate the minimal number of (focus+context)s from consideration to yield independent (focus+context)s σ́ σ́ σ́ σ́→σ́ σ́ σ́ σ́ –Indeterminacies resolved by choosing to maximize feeding and minimizing bleeding –Allow some rules to reapply

33 Constraints in TCOT Analysis T:*C LASH : λ: two consecutive [+accent] syllables, where the first vowel is short and is flanked on both sides by an obstruent. δ: [+accent] → [-accent] in the first syllable F AITH -A CCENT : penalize changes in a syllable’s accent *L APSE -A CCENT : penalize two consecutive unaccented syllables *E XT L APSE -A CCENT : penalize three consecutive unaccented syllables (Gordon 2002)

34 Constraint Ranking In the 2 foci case: σ́σ́σ́σ́ T:*C LASH ’s most preferred candidate removes 2 accents: σσσ́ Lower ranking of *L APSE -A CCENT cannot eliminate T:*C LASH ’s preferred candidate *E XT L APSE- A CCENT >> T:*C LASH >> *L APSE- A CCENT, F AITH- A CCENT

35 Constraint Ranking In the 3 foci case: σ́σ́σ́σ́σ́ T:*C LASH ’s most preferred candidate removes 3 accents: σσσσ́σ́ But the higher ranked *E XT L APSE -A CCENT prevents this candidate from being optimal T:*C LASH ’s next-most preferred candidates remove 2 accents: σ́σσσ́σ́, σσ́σσ́σ́, σσσ́σ́σ́ The lower ranked *L APSE -A CCENT chooses σσ́σσ́σ́ *E XT L APSE- A CCENT >> T:*C LASH >> *L APSE- A CCENT, F AITH- A CCENT

36 Constraint Re-Ranking Top-ranked T:*C LASH simulates –simultaneous rule application –left-to-right rule application T:*C LASH >> *E XT L APSE- A CCENT, *L APSE- A CCENT, F AITH- A CCENT

37 Constraint Re-Ranking T:*C LASH ranked below *L APSE -A CCENT simulates –right-to-left rule application *E XT L APSE- A CCENT, *L APSE- A CCENT >> T:*C LASH >> F AITH- A CCENT

38 Typology of Rule Application Targeted Constraint: analogue of a rule *L APSE: Contrast preserving constraint over 2 units *E XT -L APSE: Contrast preserving constraint over 3 units simultaneous rule application TC >> *L APSE, *E XT -L APSE

39 Typology of Rule Application Targeted Constraint: analogue of a rule *L APSE: Contrast preserving constraint over 2 units *E XT -L APSE: Contrast preserving constraint over 3 units Anderson-style simultaneous rule application *E XT -L APSE >> TC >> *L APSE

40 Typology of Rule Application Targeted Constraint: analogue of a rule *L APSE: Contrast preserving constraint over 2 units *E XT -L APSE: Contrast preserving constraint over 3 units left-to-right rule application –when δ targets the left e.g. X → Y/__ X TC >> *L APSE, *E XT -L APSE –when δ targets the righte.g. X → Y/ X __ *L APSE, *E XT -L APSE >> TC

41 Typology of Rule Application Targeted Constraint: analogue of a rule *L APSE: Contrast preserving constraint over 2 units *E XT -L APSE: Contrast preserving constraint over 3 units right-to-left rule application –when δ targets the left e.g. X → Y/__ X *L APSE, *E XT -L APSE >> TC –when δ targets the righte.g. X → Y/ X __ TC >> *L APSE, *E XT -L APSE

42 Concluding Remarks TCOT vs. Classic OT TCOT provides analyses for opacity effects TCOT avoids predicting unattested non- local interaction

43 Concluding Remarks TCOT vs. Rule-based phonology In rule-based phonology, Acoma accent loss requires Anderson’s modification In TCOT, re-ranking of constraints predicts –Acoma pattern –apparent L  R, R  L, and simultaneous rule application

44 Concluding Remarks How should we view TCOT? As a mixed model of rule-constraint interaction: a formalization of pre-OT suggestions (Paradis 1988, Myers 1991)

45 Acknowledgements Colin Wilson Luigi Burzio, Sara Finley, Bob Frank, Gaja Jarosz, Paul Smolensky

46 Bibliography Anderson, S The Organization of Phonology. New York: Academic Press. Gordon, M A factorial typology of quantity insensitive stress, Natural Language and Linguistic Theory 20, Kager, R Optimality Theory. Cambridge: Cambridge University Press. Kisseberth, C On the functional unity of phonological rules. Linguistic Inquiry 1: McCarthy, J On targeted constraints and cluster simplification. Phonology 19, Miller, W Acoma grammar and texts. (University of California Publications in Linguistics No. 40.) Berkeley: University of California Press. Myers, S Persistent Rules. Linguistic Inquiry. 22:

47 Paradis, C On Constraints and Repair Strategies. The Linguistic Review 6: Prince, A. and P. Smolensky. 1993/2004. Optimality Theory: Constraint interaction in Generative Grammar. Technical report, Rutgers University and University of Colorado at Boulder, Revised version published by Blackwell, Steriade, D The Phonology of Perceptibility Effects: the P- map and its Consequences for Constraint Organization. Unpublished manuscript, MIT. Wilson, C Consonant Cluster Neutralisation and Targeted Constraints. Phonology 18: Wilson, C. (in prep). Analyzing unbounded spreading with constraints: marks, targets, and derivations. Ms. University of California, Los Angeles.