Gestural overlap and self-organizing phonological contrasts Contrast in Phonology, University of Toronto May 3-5, 2002 Alexei Kochetov Haskins Laboratories/ Yale University
Thanks to The Project on Contrast in Phonology –SSHRC grants ( and ) to Elan Dresher and Keren Rice, University of Toronto –
Introduction Restrictions on phonological contrasts –Backness and rounding in high vowels –Secondary articulations in consonants Account: –These markedness effects emerge from low- level speaker-listener/learner interactions –The crucial role of production and perception of contrasts
Phonological contrasts Focus: –Contrasts in high vowels Front/back, rounded/unrounded Inventories /i y u/, /i y u/, or /i u/ –Contrasts in consonant secondary articulations Palatalized vs. non-palatalized: /C C ( /w) / Labialized vs. non-labialized: /C w C (w/ ) / Velarized vs. non-velarized: /C C ( ) /
Observations UPSID Database (Maddieson & Precoda 1990) –451 languages
Observations Languages with multiple vowel contrasts avoid distinctions in secondary consonant articulations –e.g. /y/ but */C j /(C = plosive; 2 exceptions)
Observations Languages with distinctive secondary articulation contrasts tend to avoid multiple vowel contrasts, particularly distinctions in rounding/backness –e.g. /C w / but */y/ (C = plosive; 1 exception)
Observations Inventories of languages of Northern and Eastern Europe –37 languages (Celtic, Germanic, Baltic, Uralic and Turkic)
Observations Karelian Faroese Icelandic Nenets Saami Mordva Chuvash Mari Tatar Bashkir
Question Why are these contrasts incompatible? ?
Explanation Approach 1 –These markedness effects are pre-specified in Universal Grammar Harmonic rankings of constraints (Optimality Theory; Prince & Smolensky 1993) Phonological representations
Approach 2 –These markedness effects arise due to lower- level factors -- limitations on production and perception –Work in phonology and phonetics: Browman & Goldstein 1986, 2002; Ohala 1981; Hume & Johnson 2001, Pierrehumbert, Beckman, & Ladd 2001, among others Cf. Jackendoff 2002 on markedness in general Explanation
Approach 2 –These markedness effects arise due to lower- level factors -- limitations on production and perception –Work in phonology and phonetics: Browman & Goldstein 1986, 2002; Ohala 1981; Hume & Johnson 2001, Pierrehumbert, Beckman, & Ladd 2001, among others Cf. Jackendoff 2002 on markedness in general
Explanation Approach 2 –Self-organization, or spontaneous emergence of order (see e.g., Kauffman 1995) dynamic systems AI and ALife (see e.g., Pfeifer & Scheier 2001)
Self-organization From www. swarm.org Simple local interaction Spontaneous emergence of order
Self-organization and phonology Phonological structure Speaker-listener interactions
Self-organization and phonology Markedness effects Unmarked: –stable with respect to production and/or perception, and/or higher-level processing –An equilibrium position Marked: –unstable with respect to production, and/or perception, and and/or higher-level processing –A non-equilibrium position
Simulation Speaker-listener/learner interactions Autonomous agents –Cf. Browman & Goldstein 1999, de Boer 2000, Lieberman 2000, Harrison, Dras & Kapicioglu 2002
A hypothetical language Language X Inventory: –{i y u} –{C C C C } Lexicon: –C 1 VC 2 words, where C 1 = C 2 –16 items
Language X Lexical items
Speaker-listener interactions Agent AAgent B * From *
Speaker-listener interactions Agent AAgent B
Speaker-listener interactions Agent AAgent B
Speaker-listener interactions Agent AAgent B
Speaker-listener interactions Agent AAgent B
Speaker-listener interactions Agent AAgent B
Speaker-listener interactions Agent AAgent B
Speaker-listener interactions Agent AAgent B
Speaker-listener interactions Agent AAgent B
Speaker-listener interactions Agent AAgent B
Speaker-listener interactions Agent AAgent B
Speaker-listener interactions Agent AAgent B
Speaker-listener interactions Agent AAgent B
Production Articulatory synthesizer (Maeda 1989, Vallée1994) Articulatory gestures (targets) Vectors of numbers between 0 and 1 –Backness [0... 1] –Height [0... 1] –Rounding [0... 1]
Production Rounding [1]
Production Rounding [0.75]
Production Rounding [0.5]
Production Rounding [0.25]
Production Rounding [0]
Production Vowels Backness Height Rounding i = [01 0 ]
Production Vowels Backness Height Rounding y = [01 1 ]
Production Consonants (secondary articulation) Backness Height Rounding C = [01 0 ]
Production Consonants (secondary articulation) Backness Height Rounding C = [01 1 ]
Production Words –Matrices of numbers between 0 and 1 e.g., C uC C uC
Production Words: sequences of gestures overlapping in time C V C
Production: Gestural overlap Gestures have conflicting targets Physical limits on how well targets can be attained An “undershoot” of at least one of the gestures (Lindblom 1963) Stiffness (GEST, Computational gestural model; Browman & Goldstein 1990)
Production: Gestural overlap Stiffness, k C = 1, k V = 1; No reduction; physically impossible C V C
Production: Gestural overlap Stiffness, k C = 1, k V = 0.75 Vowel gesture is reduced C V C
Production: Gestural overlap Evidence: –In languages with secondary articulation vowels are strongly affected by the secondary articulation quality of neighboring consonants –Russian (Bolla 1981, Kochetov 2001) –Irish (Ó Dochartaigh 1992 ) –Marshallese (Choi 1992)
Production: Gestural overlap Stiffness, k C = 0.75, k V = 1 Consonant gestures (secondary articulation) are reduced C V C
Production: Gestural overlap Evidence: –In languages with multiple backness and rounding contrasts consonants are often allophonically palatalized and velarized/labialized –Turkic languages (Comrie 1981)
Simulation Item: C uC Case 1: Vowel gesture is reduced –k C = 1, k V = 0.5 Case 2: Consonant gestures are reduced –k C = 0.5, k V = 1
Case 1 k C = 1, k V =.5 Input: C uC Output: C C or C yC
Case 2 k C = 0.5, k V = 1 Input: C uC Output: C uC or C uC
Case 1: Lexicon and grammar Agent A
Case 1: Lexicon and grammar Agent B
Case 1: Lexicon and grammar Agent B
Case 1: Lexicon and grammar Agent B
Case 1: Lexicon and grammar Default grammar 1: –limited vowel contrasts (front vs. back) –multiple consonant contrasts in secondary articulation (restricted in distribution)
Case 2: Lexicon and grammar Agent A
Case 1: Lexicon and grammar Agent B
Case 2: Lexicon and grammar Agent B
Case 2: Lexicon and grammar Agent B
Lexicon and grammar Default grammar 2: –multiple vowel contrasts (restricted in distribution) –limited consonant contrasts in secondary articulation (front vs. back) –consonants realizations are often close to neutral (non-palatalized, non-labialized, etc.)
Lexicon and grammar Grammar 0: unstable Grammar 1:more stable Grammar 2:more stable
Lexicon and grammar Default grammar 3: –limited vowel contrasts –limited consonant contrasts in secondary articulation
Conclusion The incompatibility of vowel and secondary articulation contrasts emerges through speaker-listener/learner interactions –Unstable (marked) Stable (unmarked) No reference to pre-specified “knowledge” of markedness
Limitations The simulation does not explain certain segmental markedness effects –e.g. /y/ is more marked than /i/ – / / is more marked than /u/ Markedness is a by-product of multiple factors
Further directions Implementation: additional factors –Other sequences, primary place of articulation –More realistic production and perception –More complex generalizations across the Lexicon and Grammar –Higher-level processing: morphological structure and alternations –Multiple agents: speakers/listeners Other phonological contrasts
Further directions Towards a better understanding of the phonological Grammar and markedness END