1 Introduction

Investigators have uncovered evidence for phonological learning biases: biases inherent in learners that favor certain language phonologies over others (e.g., Wilson 2006, Finley 2012, Moreton and Pater 2012, Hayes and White 2013, McMullin and Hansson 2014, White 2014). To what extent can a learning bias be defied in language? This question bears directly on the theory of phonological learning, as it addresses the limits of learner capability.

A growing family of findings suggests that learners tend to favor phonological constraints that are morphosyntactically general—that is, obeyed by at least several morphemes, or in multiple or all grammatical contexts. That phonological alternations are typically corroborated by the phonotactics of a given language was observed as early as Chomsky and Halle 1968 and Kenstowicz and Kisseberth 1977, but the generalizing tendency just mentioned has also been observed in a number of recent corpus studies. Martin (2007, 2011), Shih and Zuraw (2018), and Breiss and Hayes (2019) observe cases of grammatical “leaking,” in which strong phonotactic restrictions tend to manifest across word boundaries or compound boundaries, or affect the choice between grammatical constructions. Chong (2017) found that certain alternations purported to be apparent derived-environment effects are just that—merely apparent. For example, though Korean t-palatalization triggered by high front vocoids was previously proposed to constitute a derived-environment effect because [ti]-sequences exist in some roots, Chong showed that such sequences are highly underattested in the Korean lexicon. Generalization effects have also been borne out in artificial-language-learning experiments: Myers and Padgett (2014) found that participants generalize a phrase-final devoicing pattern to the word-final domain without exposure to unambiguous evidence; Chong (2017) found that participants more readily learned a suffixal harmony alternation when they were exposed to higher rates of root harmony, corroborating proposals that phonotactic generalizations assist in acquiring alternations (e.g., Tesar and Prince 2003, Hayes 2004, Jarosz 2006).

In light of these findings, Martin (2011) and Chong (2017) propose learning models whereby whenever the learner weights positively a structure-specific constraint (e.g., applying only across a suffix boundary), it gives positive weight to an analogous structure-insensitive constraint, leading to the generalizing tendency. If there were to exist an alternation that applies consistently in a constrained morphosyntactic context without even an analogous statistical tendency in phonotactics to accompany it, then that would complicate our understanding of learners’ preference for morphosyntactically general patterns, as it would suggest that wholly structure-specific alternations could be learned.

This squib presents one such alternation. Malagasy displays backness dissimilation, an alternation that has persisted across multiple generations that sends a back vowel to front in the presence of a nearby back vowel. The process applies very consistently to the passive imperative suffix, -u, and is blocked by intervening front vowels—a behavior typical of dissimilation—suggesting the working of an OCP (Obligatory Contour Principle) constraint. But -u is the only affix in the language that undergoes dissimilation, and is the only suffix even eligible to undergo it. Moreover, stems in the lexicon show no preference for dissimilation whatsoever; they in fact display a modest but highly significant opposing preference for harmony. This suggests that Malagasy learners induce a morphologically specific OCP constraint—specific either to -u alone or to the suffix domain as a whole—without the need for a corroborating phonotactic trend. These findings suggest that no degree of morphosyntactic generality is a necessary condition for learning. Though learners might be biased toward acquiring grammatically general constraints, the Malagasy system suggests that they are capable of overriding it. I present this system below, and discuss the problems it poses for a theory in which learners favor grammatically general constraints.

2 Backness Dissimilation Applying to the Passive Imperative Suffix

Unless otherwise specified, the data below come from the Malagasy Dictionary and Encyclopedia of Madagascar (hereafter MDEM; malagasyword.org; de La Beaujardière 2004), an annotated online corpus containing approximately 92,000 Malagasy words. The Malagasy vowel inventory is composed of [i e a u] (Parker 1883, MDEM 2004). There are four suffixes: the passive suffixes -ina and -ana, the active imperative suffix -a, and the passive imperative suffix -u (Parker 1883, Richardson 1885).1

The passive imperative suffix conditionally undergoes backness dissimilation (Parker 1883, Zymet 2015): underlying -u (1a–b) surfaces as -i after stems containing u (2a–d) unless a front vowel intervenes (3a–b). The alternation conforms to patterns driven by the OCP (Leben 1973, Goldsmith 1976, et seq.).

(1) Underlying -u

  • /bata+u/ [bata-u] ‘lift-PASS.IMP

  • b. /sava+u/ [sava-u] ‘inspect-PASS.IMP

(2) Backness dissimilation

  • /babu+u/ [babu-i] ‘plunder-PASS.IMP

  • /tuv+u/ [tuv-i] ‘fulfill-PASS.IMP

  • /suav+u/ [suav-i] ‘bless-PASS.IMP

  • /undan+u/ [undan-i] ‘bolster-PASS.IMP

(3) Blocking by front vowels

  • /turi+u/ [turi-u] ‘preach-PASS.IMP

  • /fules+u/ [fules-u] ‘thread-PASS.IMP

I extracted from MDEM 5,035 words with the passive imperative suffix. The counts in table 1 show that dissimilation is triggered by the presence of stem-internal u, applies regularly when the trigger is local and semiregularly across a, and is regularly blocked by front vowels.

Table 1

Counts for Malagasy backness dissimilation

Context (ignoring consonants)-u-iDissim. rateExample
No trigger 2,649 0.0% bata-u 
Adjacent trigger 1,365 99.6% babu-i 
Intervening a 272 245 47.3% tuda-i 
Intervening front vowel 485 0.1% turi-u 
Context (ignoring consonants)-u-iDissim. rateExample
No trigger 2,649 0.0% bata-u 
Adjacent trigger 1,365 99.6% babu-i 
Intervening a 272 245 47.3% tuda-i 
Intervening front vowel 485 0.1% turi-u 

Multiple lines of evidence suggest that Malagasy speakers acquire this alternation. For one, the process applies very consistently, and in particular essentially categorically in the local case: the -u allomorph of the suffix surfaces after nearly all of the 2,657 stems that lack u, while the -i allomorph surfaces after nearly all of the 1,371 stems whose last vowel is u. Moreover, dissimilation is observed across at least two generations: it was reported as early as Parker 1883, and evidence for it appears in dictionaries since then (e.g., Abinal and Malzac 1888, Rajemisa-Raolison 1985, MDEM 2004).

To further assess whether speakers extended the pattern to novel contexts, I conducted a study to determine whether dissimilation and its blocking is observed even when -u comes after loaned stems occurring in MDEM, as in (4a–l). Stems marked with (W) are indicated to be loans in the World Loanword Database (wold.clld.org; Adelaar 2009); stems marked with (Bl) are indicated to be loans in Blench 2008; and the rest are indicated to be loans in MDEM itself. The study uncovered 14 loaned stems whose last vowel is u; even in these stems, we observe categorical dissimilation when the vowel closest to the suffix is a trigger, reflecting the pattern observed across the corpus. The study uncovered 5 stems whose second-to-last vowel is u and whose last vowel is a. On the whole, these words vary in whether they undergo dissimilation, and therefore also reflect the corpus pattern. Finally, the study uncovered 5 stems with a triggering u separated from the suffix by front vowel(s); across all cases, we observe categorical blocking, again following the corpus pattern. These data suggest that speakers consistently extended the dissimilatory patterns observed in the corpus when they encountered eligible novel words. I provide some examples in (4); for the full set of loanwords, see the online appendix (https://www.mitpressjournals.org/doi/suppl/10.1162/ling_a_00356).

(4)

graphic

Remarkably, it would seem that the passive imperative suffix is the only affix to undergo dissimilation, and, assuming the process sends back vowels to front but not vice versa, is the only suffix even eligible to undergo it, as far as I am aware (being the only one to contain u). Even if we assume that dissimilation sends back vowels to front and vice versa, it is still not displayed by any other affix, according to an MDEM search; see the online appendix for details. If there were other evidence for a dissimilatory tendency in the grammar, we would expect to find it in phonotactics. I now turn to a corpus study of roots to assess whether this is the case.

3 A Backness Harmony Trend in Malagasy Stem Phonotactics

Surprisingly, roots display a modest but highly significant tendency toward backness harmony. MDEM gives numerous harmonic roots; for example:

(5)

  • kiri ‘small hole’

  • lufu ‘persistence’

  • gegi ‘indiscreet’

  • sarutru ‘cape’

  • tevika ‘spasm’

  • vulu ‘color’

  • uzuna ‘curse’

  • tsindri ‘compression’

  • duku ‘identity’

Counts of tier-adjacent pairs involving only front or back vowels (i, e, and u) were enumerated across 4,514 roots that were extracted from MDEM. The counts reveal no preference for disharmonic sequences in roots, as table 2 reveals. (Note that the majority of roots in the corpus are classified as nouns (2,737), adjectives (729), or adverbs (733); verbs are derived through affixation (see Keenan and Polinsky 1998).)2 There are around 350 more local harmonic sequences than local disharmonic sequences, and around 100 more nonlocal harmonic sequences than nonlocal disharmonic sequences. This makes backness dissimilation highly morphologically specific: it requires reference to the suffix domain or to the passive imperative suffix in particular, and lacks a counterpart generalization in stem phonotactics.

Table 2

Raw counts of harmonic/disharmonic sequences in roots

# harmonic# disharmonic# harmonic# disharmonic
VC0V sequencesVC0V sequencesVC0aC0V sequencesVC0aC0V sequences
Within noun roots 786 602 35 44 
Within adj. roots 185 183 27 11 
Within adv. roots 312 188 109 49 
Within interj., conj., prep. roots 96 41 24 14 
Total 1,379 1,014 205 118 
# harmonic# disharmonic# harmonic# disharmonic
VC0V sequencesVC0V sequencesVC0aC0V sequencesVC0aC0V sequences
Within noun roots 786 602 35 44 
Within adj. roots 185 183 27 11 
Within adv. roots 312 188 109 49 
Within interj., conj., prep. roots 96 41 24 14 
Total 1,379 1,014 205 118 

In fact, as we will see, Malagasy displays a significant tendency toward backness harmony in roots—these counts are unlikely to have arisen by chance alone. The observed rates of local and nonlocal harmony are 1,379/(1,379 + 1,014) = 57.3% and 205/(205 + 118) = 63.5%, respectively. We can calculate the expected local harmony rate given the frequencies of front and back vowels by extracting from roots all V1V2 sequences in which each vowel belongs to [i e u], and calculating [ p(V1 = u) × p(V2 = u)] + [ p(V1 = i or e) × p(V2 = i or e)], where (e.g.) p(V1 = u) is the number of instances of u in V1 position divided by the number of instances of i, e, and u in V1 position. The expected rate of nonlocal harmony is computed analogously over V1aV2 sequences. Doing this, we obtain 51.6% and 57.7% as expected rates of local and nonlocal harmony. Comparing the observed and expected rates, we find that observed rates (local: 57.3%; nonlocal: 63.5%) are higher than expected rates (local: 51.6%; nonlocal: 57.7%). To determine whether harmonic sequences occur significantly more than chance would predict, we can run a Monte Carlo simulation (Kessler 2001). To run a simulation for local vowel sequences, we gather pairs of tier-adjacent vowels belonging to [i e u], shuffle the second vowels of each pair and randomly concatenate each of them to a first vowel, calculate the new harmony rate, and then repeat 10,000 times. The simulation for nonlocal sequences (V1aV2 ) can be computed analogously. Figures 1a–b show histograms of local/ nonlocal harmony rate frequencies after the 10,000 trials.

Figure 1

Distribution of (a) local and (b) nonlocal harmony rates yielded by Monte Carlo trials

Figure 1

Distribution of (a) local and (b) nonlocal harmony rates yielded by Monte Carlo trials

For local harmony, the observed rate of 57.3% is greater than any rate yielded by 10,000 trials and is thus significantly greater than chance would predict (est. p <

graphic
.0001). For nonlocal harmony, the observed rate of 63.5% is greater than the rate yielded by 9,834 of the trials and is thus significantly above chance as well (est. p =
graphic
= .008). The results suggest that overrepresentation is not coincidental but reflects a backness harmony preference in phonotactics. Note that there exists some evidence of a harmony alternation as well. The -in-/-un- infix is used to create passive verbs ([sava] ‘clear’, [s-in-ava] ‘clear-PASS’; [fidi] ‘choice’, [f-in-idi] ‘choice-PASS’). MDEM gives 288 words with -in- and 14 with -un-. -in- can surface before any vowel, and in particular surfaces before u in 56 forms. But in the 14 forms with -un-, the following vowel is always u ([buri] ‘round’, [b-un-uri] = ‘round-PASS’), suggesting that -un- is selected to satisfy a (weak) harmony drive.3 That an infix can harmonize while another suffix dissimilates is reminiscent of Yucatec Maya, in which two suffixes harmonize for backness and height, but another dissimilates for backness, and yet another dissimilates for backness and height (Blair 1964; see Krämer 2001 for an account). Altogether, these cases suggest that contradictory markedness preferences can target different morphemes or domains.

To summarize, Malagasy backness dissimilation applies consistently to the passive imperative suffix and displays blocking behavior typical of the OCP. Roots, however, show no dissimilatory tendency; rather, they show a modest but highly significant harmony preference.

4 Discussion and Potential Analytical Directions

The Malagasy system complicates the picture of how a morphosyntactic generality bias in phonological learning should be modeled. Martin (2011) finds that phonotactic generalizations can “leak” into the cross-boundary domain: in Navajo sibilant harmony and English geminate avoidance, a categorical generalization within roots is mirrored by a statistical tendency across compound boundaries. Moreover, Chong’s (2017) corpus study of Korean palatalization and Turkish velar deletion, alternations previously purported to constitute derived-environment effects, suggests that the effect is merely apparent in that the alternations are either accompanied by a statistically significant phonotactic generalization or not productive. To account for these facts, Martin and Chong propose learning models that cannot acquire wholly structure-specific principles—in particular, MaxEnt learning systems whereby whenever the learner weights positively a structure-specific constraint (e.g., applying only across a suffix boundary), it gives positive weight to an analogous domain-general constraint, leading over time to generalization across domains. Perhaps a variety of morphophonological frameworks that do not encode a generalizing bias would be able to capture the data (e.g., stratal approaches, indexation-based approaches, or cophonologies); but models that cannot learn an alternation without phonotactic evidence, or that are biased to generalize the driving force for the alternation across domains, cannot obviously be applied to Malagasy, at least without further elaboration.

Here, I briefly speculate on a few potential solutions to the problem. First, perhaps when the passive imperative suffix was adopted in Malagasy, dissimilation arose to distinguish the suffix boundary (Robert Daland and Keren Rice, pers. comm.)—a drive that conflicts with the generalizing bias. Another possibility is to say that in Malagasy, the OCP targets the passive imperative suffix -u rather than the entire suffix domain, and that in general any particular morpheme can depart from typical phonological behavior in a language as long as whole domains respect the generality property to some degree. For example, palatalization is a productive alternation in Korean despite the existence of roots with [ti]-sequences (Kiparsky 1973, 1993, Iverson and Wheeler 1988); but Chong’s (2017) study shows that such roots are underattested in the language. Korean therefore displays the morphosyntactic generality property for the most part, with only some roots deviating from typical behavior. Of course, we cannot be certain that the OCP targets the passive imperative exclusively in Malagasy: it may be that the OCP scopes over the whole domain, with the one suffix violating it, -u, undergoing dissimilation. Alternatively, it could be that a generalizing bias applies even in the Malagasy case, but that Malagasy learners make use of a phonotactic harmony constraint that counteracts leaking of dissimilation into roots (see Zymet 2018 for a MaxEnt model involving this). Backness harmony and dissimilation are observed crosslinguistically (e.g., Parker 1883, Esztergár 1971, Campbell 1977, Clements and Sezer 1982, Itô 1984, Harrison 1999), and the two can even apply to different suffixes in the same language (as in both Malagasy and Yucatec Maya; see Blair 1964, Krämer 2001).4 Finally, a reviewer suggests that nonlocal (dis)harmony may inherently require domains, and therefore one might expect that drastic departure from language-wide phonotactics is possible only in such cases. A broader understanding of the typology of nonlocal restrictions—in particular, whether structure-insensitive restrictions exist—should eventually address this possibility.

5 Conclusion

Several findings now suggest that learners tend to favor morphosyntactically general phonological constraints. This squib argues that this bias, if it exists, can be overridden. Malagasy backness dissimilation applies very consistently to the passive imperative suffix -u and displays blocking behavior typical of the OCP. But -u is the only affix in the grammar that undergoes it, and is the only suffix even eligible to undergo it. Stems, on the other hand, display a modest but significant harmony trend. This suggests that Malagasy learners induce a morphologically specific OCP constraint—specific either to -u or to the suffix domain as a whole—without the need for a corroborating phonotactic trend. These findings suggest that no degree of morphosyntactic generality is a necessary condition for learning. Though learners might favor grammatically general constraints, the Malagasy system suggests that they are capable of overriding this bias.

Notes

1MDEM specifies that -u can attach to verbs to form relative imperatives, in addition to passive imperatives. I nevertheless follow Parker (1883) and Manaster-Ramer (1995), among others, in calling it a passive imperative suffix.

2 Some words displaying reduplication (see Lin 2005) were classified as roots in the corpus; in these cases, only the root involved in reduplication contributed to the counts, rather than the reduplicated stem as a whole. A conference reviewer points out that there could exist productive pseudoreduplication, with the first syllable being a copy of the second, potentially inflating the harmony rate. The corpus revealed that only 115 of the 4,514 roots have matching first and second syllables, and that only 64 begin with a front or back vowel (e.g., [diditra] ‘twisting’, [vuvuka] ‘dust’). It is not at all obvious that the language possesses pseudoreduplication, considering how low this count is.

3 An anonymous reviewer points out that these data suggest that infixes are treated as part of the stem in a way that other affixes are not. It may be that only prefixes, suffixes, and other morphemes located at the edge depart from language-wide phonotactics.

4 A reviewer points out that, if the learner can entertain dissimilation in one domain and harmony in another, then we should find a system in which harmony applies across a boundary, with roots showing a dissimilatory tendency. Whether such a system exists is a question left to future investigation.

Acknowledgments

For invaluable input, thanks to Bruce Hayes, Kie Zuraw, Adam Chong, Robert Daland, Michael Lefkowitz, Keren Rice, two anonymous Linguistic Inquiry reviewers, the Squibs and Discussion editors, and audiences at BLS43 and WCCFL35. All errors are my own.

References

References
Abinal,
Antoine
, and
Victorin
Malzac
.
1888
.
Dictionnaire Malgache-Français
.
Tananarive
:
Imprimerie de la Mission Catholique
.
Adelaar,
Alexander
.
2009
. Malagasy vocabulary. In
World loanword database
, ed. by
Martin
Haspelmath
and
Uri
Tadmor
.
Leipzig
:
Max Planck Institute for Evolutionary Anthropology
.
Blair,
Robert
.
1964
.
Yucatec Maya noun and verb morpho-syntax
.
Doctoral dissertation, Indiana University, Bloomington
.
Blench,
Roger
.
2008
. The Austronesians in Madagascar and their inter-action with the Bantu of the East African coast: Surveying the linguistic evidence for domestic and translocated animals. In
Studies in Philippine languages and cultures 18
, ed. by
Sue
McQuay
,
18
43
.
Linguistic Society of the Philippines and Summer Institute of Linguistics
.
Breiss,
Canaan
, and
Bruce
Hayes
.
2019
.
Phonological markedness effects in sentence formation
.
Ms., UCLA, Los Angeles, CA.
Campbell,
Lyle
.
1977
.
Quichean linguistic prehistory
.
Berkeley, CA
:
University of California Press
.
Chomsky,
Noam
, and
Morris
Halle
.
1968
.
The sound pattern of English
.
New York
:
Harper and Row
.
Chong,
Adam J
.
2017
.
On the relation between phonotactic learning and alternation learning
.
Doctoral dissertation, UCLA, Los Angeles, CA
.
Clements,
George
, and
Engin
Sezer
.
1982
. Vowel and consonant disharmony in Turkish. In
Structure of phonological representations (part II)
, ed. by
Harry
van der Hulst
and
Norval
Smith
,
213
255
.
Dordrecht
:
Foris
.
de La Beaujardière,
Jean-Marie
.
2004
.
Malagasy dictionary and encyclopedia of Madagascar
. malagasyword.org.
Esztergár,
Marianne
.
1971
.
A generative phonology of nouns and vowel harmony in Hungarian
.
Doctoral dissertation, University of California, San Diego
.
Finley,
Sara
.
2012
.
Typological asymmetries in round vowel harmony: Support from artificial grammar learning
.
Language and Cognitive Processes
27
:
1550
1562
.
Goldsmith,
John
.
1976
.
Autosegmental phonology
.
New York
:
Garland
.
Harrison,
K. David
.
1999
. Vowel harmony and disharmony in Tuvan and Tofa. In
Proceedings of Generative Linguistics in the Old World (GLOW) in Asia 2
, ed. by
Mamoru
Saito
,
Yasuaki
Abe
,
Hiroshi
Aoyagi
,
Masatake
Arimoto
,
Keiko
Murasugi
, and
Tatsuyo
Suzuki
,
115
130
.
Nagoya
:
Nanzan University
.
Hayes,
Bruce
.
2004
. Phonological acquisition in Optimality Theory: The early stages. In
Fixing priorities: Constraints in phonological acquisition
, ed. by
René
Kager
,
Joe
Pater
, and
Wim
Zonneveld
,
158
203
.
Cambridge
:
Cambridge University Press
.
Hayes,
Bruce
, and
James
White
.
2013
.
Phonological naturalness and phonotactic learning
.
Linguistic Inquiry
44
:
45
75
.
Itô,
Junko
.
1984
.
Melodic dissimilation in Ainu
.
Linguistic Inquiry
15
:
505
513
.
Iverson,
Gregory K.
, and
Deirdre W.
Wheeler
.
1988
. Blocking and the Elsewhere Condition. In
Theoretical morphology: Approaches in modern linguistics
, ed. by
Michael
Hammond
and
Michael
Noonan
,
325
338
.
San Diego, CA
:
Academic Press
.
Jarosz,
Gaja
.
2006
.
Rich lexicons and restrictive grammars: Maximum likelihood learning in Optimality Theory
.
Doctoral dissertation, Johns Hopkins University, Baltimore, MD
.
Keenan,
Edward
, and
Maria
Polinsky
.
1998
. Malagasy morphology. In
The handbook of morphology
, ed. by
Andrew
Spencer
and
Arnold M.
Zwicky
,
563
623
.
Oxford
:
Oxford University Press
.
Kenstowicz,
Michael
, and
Charles
Kisseberth
.
1977
.
Topics in phonological theory
.
New York
:
Academic Press
.
Kessler,
Brett
.
2001
.
The significance of wordlists
.
Cambridge, MA
:
MIT Press
.
Kiparsky,
Paul
.
1973
. Phonological representations. In
Three dimensions of linguistic theory
, ed. by
Osamu
Fujimura
,
1
135
.
Tokyo
:
TEC
.
Kiparsky,
Paul
.
1993
. Blocking in nonderived environments. In
Phonetics and phonology 4: Studies in lexical phonology
, ed. by
Ellen
Kaisse
and
Sharon
Hargus
,
277
313
.
San Diego, CA
:
Academic Press
.
Krämer,
Martin
.
2001
.
Yucatec Maya vowel alternations: Harmony as syntagmatic identity
.
Zeitschrift für Sprachwissenschaft
20
:
175
217
.
Leben,
Will
.
1973
.
Suprasegmental phonology
.
New York
:
Garland
.
Lin,
Ying
.
2005
. Two perspectives on Malagasy reduplication: Derivational and OT analyses. In
UCLA working papers in linguistics 12
, ed. by
Jeffrey
Heinz
and
Dimitris
Ntelitheos
,
255
271
.
Los Angeles, CA
:
UCLA, Department of Linguistics
.
Manaster-Ramer,
Alexis
.
1995
.
On the subject of Malagasy imperatives
.
Oceanic Linguistics
34
:
203
210
.
Martin,
Andrew
.
2007
.
The evolving lexicon
.
Doctoral dissertation, UCLA, Los Angeles, CA
.
Martin,
Andrew
.
2011
.
Grammars leak: Modeling how phonotactic generalizations interact within the grammar
.
Language
87
:
751
770
.
McMullin,
Kevin
, and
Gunnar Ólafur
Hansson
.
2014
. Locality in long-distance phonotactics: Evidence for modular learning. In
NELS 44: Proceedings of the 44th Meeting of the North East Linguistic Society
, ed. by
Jyoti
Iyer
and
Leland
Kusmer
,
1
14
.
Amherst
:
University of Massachusetts, Graduate Linguistic Student Association
.
Moreton,
Elliott
, and
Joe
Pater
.
2012
.
Structure and substance in artificial-phonology learning
.
Part 1: Structure. Language and Linguistics Compass
6
:
686
701
.
Myers,
Scott
, and
Jaye
Padgett
.
2014
.
Domain generalization in artificial language learning
.
Phonology
31
:
399
433
.
Parker,
George Williams
.
1883
.
A concise grammar of the Malagasy language
.
London
:
Trübner
.
Rajemisa-Raolison,
Régis
.
1985
.
Rakibolana Malagasy
.
Antananarivo
:
Ambozontany
.
Richardson,
James
.
1885
.
A new Malagasy-English dictionary
.
Antananarivo
:
The London Missionary Society
.
Shih,
Stephanie
, and
Kie
Zuraw
.
2018
.
Phonological conditions on variable adjective-noun word order in Tagalog
.
Phonological Data and Analysis, an online section of Language
94
:
e317
e352
.
Tesar,
Bruce
, and
Alan
Prince
.
2003
. Using phonotactics to learn phonological alternations. In
CLS 39-2: The Panels. Papers from the 39th Annual Meeting of the Chicago Linguistic Society
, ed. by
Jonathan E.
Cihlar
,
Amy L.
Franklin
,
David W.
Kaiser
, and
Irene
Kimbara
,
209
237
.
Chicago
:
University of Chicago, Chicago Lingusitic Society
.
White,
James
.
2014
.
Evidence for a learning bias against saltatory phonological alternations
.
Cognition
130
:
96
115
.
Wilson,
Colin
.
2006
.
Learning phonology with substantive bias: An experimental and computational study of velar palatalization
.
Cognitive Science
30
:
945
982
.
Zymet,
Jesse
.
2015
. Distance-based decay in long-distance phonological processes. In
Proceedings of the 32nd West Coast Conference on Formal Linguistics
, ed. by
Ulrike
Steindl
,
Thomas
Borer
,
Huilin
Fang
,
Alfredo García
Pardo
,
Peter
Guekguezian
,
Brian
Hsu
,
Charlie
O’Hara
, and
Iris Chuoying
Ouyang
,
72
81
.
Somerville, MA
:
Cascadilla Press
.
Zymet,
Jesse
.
2018
. Contradictory markedness preferences across morphological domains. In
Proceedings of the 35th West Coast Conference on Formal Linguistics
, ed. by
Wm. G.
Bennett
,
Lindsay
Hracs
, and
Dennis Ryan
Storoshenko
,
479
488
.
Somerville, MA
:
Cascadilla Proceedings Project
.

Supplementary data