Reduplication-sensitive phonology is regular Response to Hayes & - - PowerPoint PPT Presentation

Reduplication-sensitive phonology is regular

Response to Hayes & Jo 2019 ms. Hossep Dolatian, Ayla Karakas, Jeffrey Heinz

Stony Brook University

October 22, 2020

1

Introduction

Table of Contents

1

Introduction

2

Pseudo-reduplication

3

Balinese with boundaries

4

Computation of copying

5

Conclusion

2

Introduction

What is at Issue

Heinz and colleagues have been arguing that phonological patterns (phonotactics and transformations) are not only regular, but actually belong to specific subregular regions (Heinz 2007, et seq.) Hayes and Jo (HJ) draw attention to a phonotactic pattern in Balinese and argue that it is not even regular. Consequently, the subregular hypothesis is wrong. Today we provide an alternative analysis of the Balinese pattern, one that allows morphology to transfer information to phonology via boundaries ∼. A consequence of this traditional modularization of the grammar is that the phonotactics of Balinese is regular.

3

Introduction

Is phonology regular?

Regular language: definable with FSAs, regex, MSO(<) Rational relation: definable with 1-way FSTs Regular relation: definable with 2-way FSTs

1(Johnson, 1972; Kaplan and Kay, 1994; Chandlee, 2014; Chandlee and Heinz, 2012) 4

Introduction

Is phonology regular?

Regular language: definable with FSAs, regex, MSO(<) Rational relation: definable with 1-way FSTs Regular relation: definable with 2-way FSTs What about M-Phono?1

Most patterns are regular langauges Most processes are rational functions This includes partial reduplication Consequently phonotactic constraints are also regular

Except total reduplication Indonesian wanita→wanita∼wanita ‘woman’→‘women’

1(Johnson, 1972; Kaplan and Kay, 1994; Chandlee, 2014; Chandlee and Heinz, 2012) 4

Introduction

What’s wrong with total reduplication?

As a language: Multiple-Context-Free (MCF)2

2(Gazdar and Pullum, 1985; Seki et al., 1991; Albro, 2005; Clark and Yoshinaka, 2016) 3(Koskenniemi, 1983; Beesley and Karttunen, 2003) 5

Introduction

What’s wrong with total reduplication?

As a language: Multiple-Context-Free (MCF)2 As a function: requires 2-way FSTs (Dolatian and Heinz, 2020) Recognizing copies is harder than making copies.

2(Gazdar and Pullum, 1985; Seki et al., 1991; Albro, 2005; Clark and Yoshinaka, 2016) 3(Koskenniemi, 1983; Beesley and Karttunen, 2003) 5

Introduction

What’s wrong with total reduplication?

As a language: Multiple-Context-Free (MCF)2 As a function: requires 2-way FSTs (Dolatian and Heinz, 2020) Recognizing copies is harder than making copies. → RED-based phonology is also MCF & requires 2-way FST

2(Gazdar and Pullum, 1985; Seki et al., 1991; Albro, 2005; Clark and Yoshinaka, 2016) 3(Koskenniemi, 1983; Beesley and Karttunen, 2003) 5

Introduction

What’s wrong with total reduplication?

As a language: Multiple-Context-Free (MCF)2 As a function: requires 2-way FSTs (Dolatian and Heinz, 2020) Recognizing copies is harder than making copies. → RED-based phonology is also MCF & requires 2-way FST special tonal contours morpheme-specific rules c´

• py → c´
• py∼c`
• py

book → book∼schmook (Downing, 2003) (Inkelas and Zoll, 2005)

2(Gazdar and Pullum, 1985; Seki et al., 1991; Albro, 2005; Clark and Yoshinaka, 2016) 3(Koskenniemi, 1983; Beesley and Karttunen, 2003) 5

Introduction

What’s wrong with total reduplication?

As a language: Multiple-Context-Free (MCF)2 As a function: requires 2-way FSTs (Dolatian and Heinz, 2020) Recognizing copies is harder than making copies. → RED-based phonology is also MCF & requires 2-way FST special tonal contours morpheme-specific rules c´

• py → c´
• py∼c`
• py

book → book∼schmook (Downing, 2003) (Inkelas and Zoll, 2005) Intersection language of Red&phonology isn’t regular, but...

2(Gazdar and Pullum, 1985; Seki et al., 1991; Albro, 2005; Clark and Yoshinaka, 2016) 3(Koskenniemi, 1983; Beesley and Karttunen, 2003) 5

Introduction

What’s wrong with total reduplication?

As a language: Multiple-Context-Free (MCF)2 As a function: requires 2-way FSTs (Dolatian and Heinz, 2020) Recognizing copies is harder than making copies. → RED-based phonology is also MCF & requires 2-way FST special tonal contours morpheme-specific rules c´

• py → c´
• py∼c`
• py

book → book∼schmook (Downing, 2003) (Inkelas and Zoll, 2005) Intersection language of Red&phonology isn’t regular, but... Finite-state compromise = modulate the grammar:3

Morphology: reduplicates, applies Red morpheme boundary ∼ Phonology: reads ∼ boundary, applies constraints/rules

2(Gazdar and Pullum, 1985; Seki et al., 1991; Albro, 2005; Clark and Yoshinaka, 2016) 3(Koskenniemi, 1983; Beesley and Karttunen, 2003) 5

Introduction

Data (Balinese)

CC sequences are homorganic NT sequences ... naNkil sampun tamban *meptik *ranbap *lumtik

6

Introduction

Data (Balinese)

CC sequences are homorganic NT sequences ... naNkil sampun tamban *meptik *ranbap *lumtik ... unless the two syllables are identical:

6

Introduction

Data (Balinese)

CC sequences are homorganic NT sequences ... naNkil sampun tamban *meptik *ranbap *lumtik ... unless the two syllables are identical: patpat nemnem kenken

6

Introduction

Data (Balinese)

CC sequences are homorganic NT sequences ... naNkil sampun tamban *meptik *ranbap *lumtik ... unless the two syllables are identical: patpat nemnem kenken “Pseudo-reduplication”: RED words aren’t derived from words

6

Introduction

Data (Balinese)

CC sequences are homorganic NT sequences ... naNkil sampun tamban *meptik *ranbap *lumtik ... unless the two syllables are identical: patpat nemnem kenken “Pseudo-reduplication”: RED words aren’t derived from words *pat *nem *ken

6

Introduction

HJ’s Account

Pseudo-reduplicated forms are computationally beyond regular Pattern: Non-homorganic CC is only permitted in ww words patpat nemnem kenken Theory:

Lexicon divided into Core and Reduplicated strata MaxEnt grammar gives different weights to CodaCondition

”an adequate phonotactic analysis of Balinese must have access to the information of whether a stem is pseudoreduplicated or not; which implies that the phonotactic assessment in general must include the capacity to detect copied strings.” {ww ∶ w ∈ Σ∗} is not regular → Balinese phonotactics isn’t regular

7

Introduction

Rebuttal

Inherently reduplicated forms in Balinese (and in many other cases) are instances of morphological pseudo-reduplication Morphological pseudo-reduplication puts morph boundaries into the representation. pat∼pat nem∼nem ken∼ken

8

Introduction

Rebuttal

Inherently reduplicated forms in Balinese (and in many other cases) are instances of morphological pseudo-reduplication Morphological pseudo-reduplication puts morph boundaries into the representation. pat∼pat nem∼nem ken∼ken C∼C does not match the environment specified by *CC → *CC is a regular language HJ rely heavily on Zuraw’s theory of pseudo-reduplication as ‘aggressive’ phonological reduplication. Essentially, this counter-argument says that not all pseudo reduplication is of this ‘aggressive’ variety.

8

Introduction

Going forward

Goal: explain how our encoding is plausible pat∼pat nem∼nem ken∼ken

9

Introduction

Going forward

Goal: explain how our encoding is plausible pat∼pat nem∼nem ken∼ken

9

Introduction

Going forward

Goal: explain how our encoding is plausible pat∼pat nem∼nem ken∼ken Note:

9

Introduction

Going forward

Goal: explain how our encoding is plausible pat∼pat nem∼nem ken∼ken Note:

Data: We rely a lot on Clynes (1995) (Cl), a large-scale study

• f Balinese

HJ talk more about statistics and metrics for treating phonology as regular (appendix)

9

Pseudo-reduplication

Table of Contents

1

Introduction

2

Pseudo-reduplication

3

Balinese with boundaries

4

Computation of copying

5

Conclusion

10

Pseudo-reduplication

Types of reduplication

Reduplicative constructs based on lexical meaning: Red (“meaningful”) reduplicated ww is derived from lexical item w Indonesian wanita→wanita∼wanita ‘woman’→‘women’

11

Pseudo-reduplication

Types of reduplication

Reduplicative constructs based on lexical meaning: Red (“meaningful”) reduplicated ww is derived from lexical item w Indonesian wanita→wanita∼wanita ‘woman’→‘women’ Pseudo-Red (“meaningless”) no lexical item w English orangutan → orangutang, hodge podge Chamorro lamlam ‘glare’

11

Pseudo-reduplication

Types of reduplication

Reduplicative constructs based on lexical meaning: Red (“meaningful”) reduplicated ww is derived from lexical item w Indonesian wanita→wanita∼wanita ‘woman’→‘women’ Pseudo-Red (“meaningless”) no lexical item w English orangutan → orangutang, hodge podge Chamorro lamlam ‘glare’ We distinguish two types of pseudo-Red (HJ & Zuraw (2002) do not make this distinction)

11

Pseudo-reduplication

Types of pseudo-reduplication

Phonological pseudo-Red a single morph becomes internally repetitive due to phonological changes (Zuraw, 2002; Inkelas and Zoll, 2005) English: orangutan → orangutang Tagalog: /kalos + in/ → [kalus-in] but /todo + in/ → [todo-hin] (see appendix)

12

Pseudo-reduplication

Types of pseudo-reduplication

Phonological pseudo-Red a single morph becomes internally repetitive due to phonological changes (Zuraw, 2002; Inkelas and Zoll, 2005) English: orangutan → orangutang Tagalog: /kalos + in/ → [kalus-in] but /todo + in/ → [todo-hin] (see appendix) Note that Inkelas and Zoll (2005) and Inkelas (2008) say these effects are bounded locally and not instances of reduplication.

12

Pseudo-reduplication

Morphological pseudo-Red a word can be morphologically decomposed to two bound, identical(-ish) morphs (Stolz et al., 2009) Chamorro: lamlam ‘glare’ English: hodge podge

13

Pseudo-reduplication

Morphological pseudo-Red a word can be morphologically decomposed to two bound, identical(-ish) morphs (Stolz et al., 2009) Chamorro: lamlam ‘glare’ English: hodge podge Computation is the same as Red

Morphology applies a reduplicative transformation (not regular, sensitive to length) Phonology references the ∼ boundary (regular)

→ Balinese exhibits morphological pseudo-reduplication

13

Balinese with boundaries

Table of Contents

1

Introduction

2

Pseudo-reduplication

3

Balinese with boundaries

4

Computation of copying

5

Conclusion

14

Balinese with boundaries

Inherent reduplication

Morphological pseudo-Red: Chamorro lamlam ‘glare’ English hodge podge

15

Balinese with boundaries

Inherent reduplication

Morphological pseudo-Red: Chamorro lamlam ‘glare’ English hodge podge Balinese:

These ‘identical’ syllables are separated by a Red boundary ∼ pat∼pat nem∼nem ken∼ken Heterorganic CC is fine across ∼ boundary

1 morpheme = 2 identical morphs

15

Balinese with boundaries

Inherent reduplication

Morphological pseudo-Red: Chamorro lamlam ‘glare’ English hodge podge Balinese:

These ‘identical’ syllables are separated by a Red boundary ∼ pat∼pat nem∼nem ken∼ken Heterorganic CC is fine across ∼ boundary

1 morpheme = 2 identical morphs Unlike morphemes, morphs don’t have to be semantically meaningful, just structurally decomposable (Aronoff, 1976) re-ceive hodge-podge in-ert

15

Balinese with boundaries

Balinese: existing analysis

Our morph analysis isn’t original Clynes (1995, 70) decomposes these words to 2 morphs = One morpheme but two copies of a morph pat (via Red) [pat∼pat] [nem∼nem] [ken∼ken] /Red - pat/ /Red - nem/ /Red - ken/

4Buckley (1997); Rose (1997) 16

Balinese with boundaries

Balinese: existing analysis

Our morph analysis isn’t original Clynes (1995, 70) decomposes these words to 2 morphs = One morpheme but two copies of a morph pat (via Red) [pat∼pat] [nem∼nem] [ken∼ken] /Red - pat/ /Red - nem/ /Red - ken/ FYI: a common solution for morphological pseudo-Red before Zuraw (2002)4 How well does this analysis work?

4Buckley (1997); Rose (1997) 16

Balinese with boundaries

Balinese: existing analysis

Our morph analysis isn’t original Clynes (1995, 70) decomposes these words to 2 morphs = One morpheme but two copies of a morph pat (via Red) [pat∼pat] [nem∼nem] [ken∼ken] /Red - pat/ /Red - nem/ /Red - ken/ FYI: a common solution for morphological pseudo-Red before Zuraw (2002)4 How well does this analysis work? Very!

4Buckley (1997); Rose (1997) 16

Balinese with boundaries

Size

Pseudo-Red can be bigger than 2 syllables: lumba∼lumba kupu∼kupu kisi∼kisi ‘porpoise’ ‘butterfly’ ‘whisper’

17

Balinese with boundaries

Size

Pseudo-Red can be bigger than 2 syllables: lumba∼lumba kupu∼kupu kisi∼kisi ‘porpoise’ ‘butterfly’ ‘whisper’ Also violate the constraint dangap∼dangap kebarat∼kebirit ‘lizard. sp.’ ‘run about wildly’ Orthography: use a hyphen if 3++ syllables (Cl:xvi) cakcak kupu-kupu kebarat-kebirit ‘chop up’ ‘butterfly’ ‘run about wildly’ Speakers perceive longer words as reduplicated, while the shorter ones are not (Cl:82)

17

Balinese with boundaries

Size

Pseudo-Red can be bigger than 2 syllables: lumba∼lumba kupu∼kupu kisi∼kisi ‘porpoise’ ‘butterfly’ ‘whisper’ Also violate the constraint dangap∼dangap kebarat∼kebirit ‘lizard. sp.’ ‘run about wildly’ Orthography: use a hyphen if 3++ syllables (Cl:xvi) cakcak kupu-kupu kebarat-kebirit ‘chop up’ ‘butterfly’ ‘run about wildly’ Speakers perceive longer words as reduplicated, while the shorter ones are not (Cl:82) All have a meaningless reduplicant and ∼ boundary dangap∼dangap /Red+dangap/ ‘lizard’

17

Balinese with boundaries

We need morphological information

Does morphology affect this constraint in general?

18

Balinese with boundaries

We need morphological information

Does morphology affect this constraint in general? Very! CC constraint is also blocked in meaningful Red (Cl:76,156) bikul ‘rat’ bikul∼bikul ‘rats’ luh ‘female’ luh∼luh ‘females’

18

Balinese with boundaries

We need morphological information

Does morphology affect this constraint in general? Very! CC constraint is also blocked in meaningful Red (Cl:76,156) bikul ‘rat’ bikul∼bikul ‘rats’ luh ‘female’ luh∼luh ‘females’ Blocked in loanwords from Sanskrit (Cl:93) ap.sari bak.ti is.tri brah.ma ‘angel’ ‘homage’ ‘female’ ‘Brahma’ → Constraint is sensitive to non-phonological information from morphology

18

Balinese with boundaries

More evidence for a boundary

Copies carry two stresses (Clynes, 1995, 171) c´ u∼c´ u g´ es∼g´ es cr´ uk∼cr´ u ‘grandchild’ ‘scratch (vt)’ ‘k.o. bird’ Some rules are sensitive to ∼ boundary, some are not, such as vowel laxing (Clynes, 1995, 100,171)5 git∼gIt gud∼gUd pruN∼prUN ’place name’ ’extremely ’chipped’ Some rules reference ∼ boundary in only small words,

C.C in disyllabic copies can’t be identical, but they can in larger copies (Clynes, 1995, 172).

*kak∼kak *kik∼kik kayak∼kayak kisik∼kisik ’lie stretched out’ ’move slightly’

5He uses cyclicity and rule ordering 19

Balinese with boundaries

Interim summary

Reduplication is not a regular language

6Clynes (1995, 91) actually argues that morphology applies after the coda restriction is turned off. 20

Balinese with boundaries

Interim summary

Reduplication is not a regular language Phonology looks at reduplication in a regular local way 6 Input: /naNkil/ /Red + pat/ Morpho pat∼pat Phono ✓ ✓

6Clynes (1995, 91) actually argues that morphology applies after the coda restriction is turned off. 20

Balinese with boundaries

Interim summary

Reduplication is not a regular language Phonology looks at reduplication in a regular local way 6 Input: /naNkil/ /Red + pat/ Morpho pat∼pat Phono ✓ ✓ → Balinese is not a case for supra-regular phonology Otherwise, the phonology of every language with morphological Red would be called non-regular

6Clynes (1995, 91) actually argues that morphology applies after the coda restriction is turned off. 20

Balinese with boundaries

Interim summary

Reduplication is not a regular language Phonology looks at reduplication in a regular local way 6 Input: /naNkil/ /Red + pat/ Morpho pat∼pat Phono ✓ ✓ → Balinese is not a case for supra-regular phonology Otherwise, the phonology of every language with morphological Red would be called non-regular ∼ syntax (CF) vs. prosody (regular) (Pierrehumbert, 1980) Crux of HJ’s argument relies on Zuraw’s analysis of pseudo-Red.

6Clynes (1995, 91) actually argues that morphology applies after the coda restriction is turned off. 20

Computation of copying

Table of Contents

1

Introduction

2

Pseudo-reduplication

3

Balinese with boundaries

4

Computation of copying

5

Conclusion

21

Computation of copying

Copying across modules

Repetition-detection is a major part of cognition (Endress et al., 2007; Alhama and Zuidema, 2019) Lots of copying-based patterns in natural language

22

Computation of copying

Copying across modules

Repetition-detection is a major part of cognition (Endress et al., 2007; Alhama and Zuidema, 2019) Lots of copying-based patterns in natural language

• 1. Morphology (whether meaningful or not)

Partial or total Red

22

Computation of copying

Copying across modules

Repetition-detection is a major part of cognition (Endress et al., 2007; Alhama and Zuidema, 2019) Lots of copying-based patterns in natural language

• 1. Morphology (whether meaningful or not)

Partial or total Red

• 2. Phonology

Copy epenthesis (Stanton and Zukoff, 2018) OCP (Reiss, 2003, 2008) Vowel harmony (Nevins, 2010) Phonological pseudo-Red (Zuraw, 2002)

22

Computation of copying

Copying across modules

Repetition-detection is a major part of cognition (Endress et al., 2007; Alhama and Zuidema, 2019) Lots of copying-based patterns in natural language

• 1. Morphology (whether meaningful or not)

Partial or total Red

• 2. Phonology

Copy epenthesis (Stanton and Zukoff, 2018) OCP (Reiss, 2003, 2008) Vowel harmony (Nevins, 2010) Phonological pseudo-Red (Zuraw, 2002)

• 3. Syntax (Kobele, 2006; Clark and Yoshinaka, 2014)

Verb copying Recursive copying in Yoruba (a2n)

22

Computation of copying

Copying across modules

Repetition-detection is a major part of cognition (Endress et al., 2007; Alhama and Zuidema, 2019) Lots of copying-based patterns in natural language

• 1. Morphology (whether meaningful or not)

Partial or total Red

• 2. Phonology

Copy epenthesis (Stanton and Zukoff, 2018) OCP (Reiss, 2003, 2008) Vowel harmony (Nevins, 2010) Phonological pseudo-Red (Zuraw, 2002)

• 3. Syntax (Kobele, 2006; Clark and Yoshinaka, 2014)

Verb copying Recursive copying in Yoruba (a2n)

• 4. Style: Alliteration & rhyme (Holtman, 1996; Yip, 1999, 2001)

Haldeman and Ehrlichman becomes Herdleman and Erdleman

22

Computation of copying

Computationally, copying is different

Computationally, copying is really different across domains7 Phonology Subregular Morphology MCFG over segments Syntax PMCFG over words Style ?

7(Stanton and Zukoff, 2018; Culy, 1985; Clark and Yoshinaka, 2014; Kobele, 2006) 23

Computation of copying

Computationally, copying is different

Computationally, copying is really different across domains7 Phonology Subregular Morphology MCFG over segments Syntax PMCFG over words Style ? → How phonology uses ‘identity’ or ‘copying’ should be different from how morphology does it (Inkelas and Zoll, 2005)

7(Stanton and Zukoff, 2018; Culy, 1985; Clark and Yoshinaka, 2014; Kobele, 2006) 23

Computation of copying

What is copying?

Formal language theory is not only about sets of strings. It is also about transformations: string-to-string relations. It is about computational laws of processing not only sequences but other structured data types. There are subtleties to copying that have not been appreciated.

Recognizing the copy language ww is more complex than computing the copy function w ↦ ww.

There are sure to be more, and we have just scratched the surface of their implications for generative grammars.

24

Conclusion

Table of Contents

1

Introduction

2

Pseudo-reduplication

3

Balinese with boundaries

4

Computation of copying

5

Conclusion

25

Conclusion

Conclusion: Happy medium

Total reduplication isn’t regular (as a language) Reduplication-sensitive phonology seems to be non-regular But once Red is given in terms of morph division, the rest of phonology is regular once again Copying is computationally special and intersects all modules

26

Guide

Thanks for listening Not sure about a question? Here’s the appendix

• 1. Copying power? [28]
• 2. More on HJ account? [29]
• 3. How to evaluate a grammar?

3.1 Generative capacity [30] 3.2 HJ’s Metric and generative capacity [31] 3.3 Don’t give up on regularity [35] 3.4 Implementation & memorization of Red [36]

• 4. More on typology of pseudo-Red

4.1 Morphological pseudo-Red [37] 4.2 Phonological pseudo-Red [38]

• 5. More on Balinese boundaries [40]

27

Copying differences

What is total reduplication?

The copy language {ww ∣ w ∈ Σ∗} is not context-free but Multi-Context-Free (MCF)8 As a function, needs 2-way FSTs (Dolatian and Heinz, 2020)

In this sense, recognizing the copy language ww is more complex than computing the copy function w ↦ ww.

8(Gazdar and Pullum, 1985; Seki et al., 1991; Albro, 2005; Clark and

Yoshinaka, 2016)

28

HJ theory

Words are divided into two strata: Core and Reduplicated *CC is formalized with CodaCondition patpat nemnem kenken HJ theory: In a MaxEnt grammar, the weight of CodaCondition in the Core and Reduplicated strata are very different. = ”an adequate phonotactic analysis of Balinese must have access to the information of whether a stem is pseudoreduplicated or not; which implies that the phonotactic assessment in general must include the capacity to detect copied strings.” → {ww ∣ w ∈ Σ∗} is not a regular language so Balinese phonotactics isn’t regular

29

Generative capacity

What is it?

Weak Generative Capacity (WGC) = the surface stringset Strong Generative Capacity (SGC) = the hidden structure

HJ conflate the two in their arguments because they don’t use the specialized term ‘SGC’

30

Generative capacity

What is it?

Weak Generative Capacity (WGC) = the surface stringset Strong Generative Capacity (SGC) = the hidden structure

HJ conflate the two in their arguments because they don’t use the specialized term ‘SGC’ SGC: partial Red requires memorization with an FSA

Everybody: that’s burdensome and doesn’t match theory HJ: so we shouldn’t say it’s regular Us: but only for SGC so should be careful (appendix)

WGC: partial Red is regular

30

How to evaluate grammars

HJ’s metric:

A good grammar is one that generalizes well If you add more examples to it, the grammar shouldn’t grow Partial reduplication adds more segments to the FSA

Precedence: Pullum (1983)’s profligacy constraint But their metric is...

too informal under-generates phonology would faithfully require super-expressive grammars like Parallel-Multi-Context-Free Grammars

31

Informality

HJ are essentially evaluating grammars for their strong generative capacity (but don’t say it) But work in SCC often leads to dead-ends because the desired structures are too circular and theory-dependent (Miller, 1999) If want to use strong metrics for generative capacity, there are

• ther more formalized systems...
• 1. Order-preservation for phonology (Chandlee and Jardine,

2019)

• 2. Order-preservation for morphology (Dolatian, 2020)
• 3. Origin semantics for reduplication (Dolatian and Heinz, 2020)
• 4. Origin semantics for phonology (Jon+Dakotah)

32

Undergeneration

HJ’s metric not only bans partial reduplication as regular, but it also bans cases where... Every time you add a segment to your inventory, you increase your machine

• 1. OCP: Add schwa between two identical segments
• 2. Anti-OCP: Delete schwa between two identical segments

Every time you add a natural class, you increase your machine

• 1. Place assimilation: assimilate nasals to following stops
• 2. Vowel harmony: all vowels are identical in every feature

33

Supraregular grammars

If you want to model any of these phonological patterns in a more compact way, you would need PMCFG (Clark and Yoshinaka, 2014)

• 1. Partial reduplication
• 2. OCP: Add schwa between two identical segments
• 3. Anti-OCP: Delete schwa between two identical segments
• 4. Place assimilation: assimilate naasls to following stops
• 5. Vowel harmony: all vowels are identical in every feature
• 6. Copy epenthesis (Stanton and Zukoff, 2018)

What can PMCFGs do? Anything in natural syntax and more

34

Counter 1: Give up on regularity?

HJ train-of-thought: if reduplication isn’t regular, then give up on subregular hypothesis But... that’s extreme

Overwhelming majority of MPhono is (sub)regular → still need to capture that tendency

Analog: what’s syntax?

• verwhelming majority of syntax is context-free (Pullum and

Gazdar, 1982), except for... Crossing dependencies (Shieber, 1985) Syntactic copying: a2n (Kobele, 2006) And a handful more (Clark and Yoshinaka, 2014)

→ But doesn’t mean we should stop using context-free grammars at all

35

Counter 2: Memorization

HJ:

Even if you push reduplication into being regular, then you have to memorize all these syllables Implication: We should separate finite-state generalizations from infinite-sized ones (Savitch, 1993)

But that’s the mathematical properties of the phenomenon itself! And besides...

• 1. State-explosion can be handled in some equivalent

implementations, e.g. FSA with bounded registers (Cohen-Sygal and Wintner, 2006)

• 2. No memorization needed with functions (Dolatian and Heinz,

2020)

36

Morphological pseudo-Red

Morphological pseudo-Red is common cross-linguistically (Stolz et al., 2009), especially in Austronesian (Blust, 2013) Morphological pseudo-Red has strong semantic connections to typical reduplication (Mattes, 2007)

Iconicity, onomatopoeia, ...

Diachrony can remove productive morphology + identity via sound changes (Blust, 2001)

37

Aggressive reduplication

We distinguish between:

• 1. Morphological pseudo-Red: Balinese
• 2. Phonological pseudo-Red or Aggrssive Reduplication:

increasing similarity of nearly-identical sequence of syllables

Zuraw (2002) illustrates phono pseudo-Red in Tagalog

Loanwords can take native suffixation and undergo final mid-raising /kalos + in/ [kalus-in] Loanwords are more likely to resist final mid-raising if the predeing vowel is mid /todo + in/ [todo-hin]

Analyzes these with Redup and coupling as a type of correspondence Data cannot be morphologically pseudo-Red because only a single morph

38

Coupling without Redup

Zuraw: The constraint Redup sets up coupling relationships which trigger identity-effects over substrings This is essentially just correspondence (Inkelas and Zoll, 2005, 221) Correspondence is commonly argued for pseudo-Red in Semitic (Rose, 1997) This correspondence is shows proximity and is local (Inkelas, 2008, 365,84)

39

Need for boundary

A lot of evidence for morphological decomposition and need for a boundary Copies carry two stresses (Clynes, 1995, 171) c´ u∼c´ u g´ es∼g´ es cr´ uk∼cr´ u ‘grandchild’ ‘scratch (vt)’ ‘k.o. bird’ Some rules are sensitive to ∼ boundary, some are not, such as vowel laxing (Clynes, 1995, 100,171)9 git∼gIt gud∼gUd pruN∼prUN ’place name’ ’extremely ’chipped’ Some rules reference ∼ boundary in only small words,

C.C in disyllabic copies can’t be identical, but they can in larger copies (Clynes, 1995, 172).

*kak∼kak *kik∼kik kayak∼kayak kisik∼kisik ’lie stretched out’ ’move slightly’

9He uses cyclicity and rule ordering 40

References

41

References

Albro, D. M. (2005). Studies in Computational Optimality Theory, with Special Reference to the Phonological System of Malagasy.

• Ph. D. thesis, University of California, Los Angeles, Los Angeles.

Alhama, R. G. and W. Zuidema (2019). A review of computational models of basic rule learning: The neural-symbolic debate and

• beyond. Psychonomic bulletin & review 26(4), 1–21.

Aronoff, M. (1976). Word formation in generative grammar. Number 1 in Linguistic Inquiry Monographs. Cambridge, MA: The MIT Press. Beesley, K. and L. Karttunen (2003). Finite-state morphology: Xerox tools and techniques. Stanford, CA: CSLI Publications. Blust, R. (2001). Reduplicated colour terms in Oceanic languages. In A. P. M. Ross and D. Tryon (Eds.), The boy from Bundaberg: Studies in Melanesian linguistics in honour of Tom Dutton, Number 514 in Pacific Linguistics, pp. 23–49. Canberra, Australia: Australian National University.

41

References

Blust, R. (2013). The Austronesian languages. Asia-Pacific Open-Access Monographs. Canberra: Asia-Pacific Linguistics. Buckley, E. (1997). Integrity and correspondence in Manam double

• reduplication. In Proceedings of NELS, Volume 28, pp. 59–67.

Chandlee, J. (2014). Strictly Local Phonological Processes. Ph. D. thesis, University of Delaware, Newark, DE. Chandlee, J. and J. Heinz (2012). Bounded copying is subsequential: Implications for metathesis and reduplication. In Proceedings of the 12th Meeting of the ACL Special Interest Group on Computational Morphology and Phonology, SIGMORPHON ’12, Montreal, Canada, pp. 42–51. Association for Computational Linguistics. Chandlee, J. and A. Jardine (2019). Quantifier-free least fixed point functions for phonology. In Proceedings of the 16th Meeting on the Mathematics of Language (MoL 16), Toronto,

• Canada. Association for Computational Linguistics.

41

References

Clark, A. and R. Yoshinaka (2014). Distributional learning of parallel multiple context-free grammars. Machine Learning 96(1-2), 5–31. Clark, A. and R. Yoshinaka (2016). Distributional learning of context-free and multiple context-free grammars. In J. Heinz and J. M. Sempere (Eds.), Topics in Grammatical Inference, pp. 143–172. Berlin, Heidelberg: Springer. Clynes, A. (1995). Topics in the phonology and morphosyntax of Balinese based on the dialect of Singaraja, North Bali. Ph. D. thesis, The Australian National University. Cohen-Sygal, Y. and S. Wintner (2006). Finite-state registered automata for non-concatenative morphology. Computational Linguistics 32(1), 49–82. Culy, C. (1985). The complexity of the vocabulary of Bambara. Linguistics and Philosophy 8, 345–351. Dolatian, H. (2020). Computational locality of cyclic phonology in

• Armenian. Ph. D. thesis, Stony Brook University.

41

References

Dolatian, H. and J. Heinz (2020). Computing and classifying reduplication with 2-way finite-state transducers. Journal of Language Modeling 8, 79–250. Downing, L. J. (2003). Compounding and tonal non-transfer in Bantu languages. Phonology 20(1), 1–42. Endress, A. D., G. Dehaene-Lambertz, and J. Mehler (2007). Perceptual constraints and the learnability of simple grammars. Cognition 105(3), 577–614. Gazdar, G. and G. K. Pullum (1985). Computationally relevant properties of natural languages and their grammars. New generation computing 3, 273–306. Holtman, A. (1996). A generative approach to rhyme: An Optimality approach. Ph. D. thesis, Universiteit Utrecht, Utrecht. Inkelas, S. (2008). The dual theory of reduplication. Linguistics 46(2), 351–401.

41

References

Inkelas, S. and C. Zoll (2005). Reduplication: Doubling in

• Morphology. Cambridge: Cambridge University Press.

Johnson, C. D. (1972). Formal aspects of phonological description. The Hague: Mouton. Kaplan, R. M. and M. Kay (1994). Regular models of phonological rule systems. Computational linguistics 20(3), 331–378. Kobele, G. M. (2006). Generating Copies: An investigation into structural identity in language and grammar. Ph. D. thesis, University of California, Los Angeles. Koskenniemi, K. (1983). Two-level morphology: A General Computational Model for Word-Form Recognition and

• Production. Ph. D. thesis, University of Helsinki.

Mattes, V. (2007). Reduplication in Bikol. Ph. D. thesis, University of Graz, Graz, Austria. Miller, P. H. (1999). Strong generative capacity: The semantics of linguistic formalism. Stanford: CSLI publications.

41

References

Nevins, A. (2010). Locality in vowel harmony. Cambridge: MIT Press. Pierrehumbert, J. B. (1980). The phonology and phonetics of English intonation. Ph. D. thesis, Massachusetts Institute of Technology. Pullum, G. K. (1983). Context-freeness and the computer processing of human languages. In 21st Annual Meeting of the Association for Computational Linguistics, pp. 1–6. Pullum, G. K. and G. Gazdar (1982). Natural languages and context-free languages. Linguistics and Philosophy 4(4), 471–504. Reiss, C. (2003). Quantification in structural descriptions: Attested and unattested patterns. The linguistic review 20(2-4), 305–338. Reiss, C. (2008). Constraining the learning path without constraints, or the OCP and NoBanana. In B. Vaux and

• A. Nevins (Eds.), Rules, Constraints, and Phonological

41

References

Phenomena, Chapter 8, pp. 252–301. Oxford: Oxford University Press. Rose, S. (1997). Multiple correspondence in reduplication. In Proceedings of the Twenty-Third Annual Meeting of the Berkeley Linguistics Society: General Session and Parasession on Pragmatics and Grammatical Structure, Berkeley, pp. 315–326. Berkeley Linguistics Society. Savitch, W. J. (1993). Why it might pay to assume that languages are infinite. Annals of Mathematics and Artificial Intelligence 8(1-2), 17–25. Seki, H., T. Matsumura, M. Fujii, and T. Kasami (1991). On multiple context-free grammars. Theoretical Computer Science 88(2), 191–229. Shieber, S. M. (1985). Evidence against the context-freeness of natural language. Linguistics and Philosophy, 333–343. Stanton, J. and S. Zukoff (2018). Prosodic identity in copy

41

• epenthesis. Natural Language & Linguistic Theory 36(2),

637–684. Stolz, T., N. Levkovych, and B. Dewein (2009). Reduplicative stem formation: A comparative look at Maltese and Chamorro. Ilsiena 1, 87–133. Yip, M. (1999). Reduplication as alliteration and rhyme. Glot International 4(8), 1–7. Yip, M. (2001). Segmental unmarkedness versus input preservation in reduplication. In L. Lombardi (Ed.), Segmental phonology in

• ptimality theory: constraints and representations, pp. 206–228.

Cambridge: Cambridge University Press. Zuraw, K. (2002). Aggressive reduplication. Phonology 19(3), 395–439.

41