Questions for the Week

1
Questions for the Week
2
Questions

• Can (all?) modifications in foreign language
  production be explained as transfer from the
  native language grammar? Does universal
  markedness play a role? Language-independent
  phonetic factors?

3
Questions, continued

• Why are some foreign structures mastered more
  quickly than other (equally new) structures?
• Japanese chiimu team
• shiifuudo seafood
• shiitibanku Citibank
• never siichibanku

4
Questions, continued

• What determines how illegal structures are
  modified?
• Christmas
• Japanese kurisumasu
• Hawaiian kalikimaki
• Maori kirihimete
• Samoan kilisimasi

5
Questions, continued

• How much of foreign language modification is a
  result of misperception? Of misproduction? And
  how can we tell?

6
For example

• Does a Japanese speaker who pronounces
  Christmas as kurisumasu actually HEAR the
  English pronunciation as kurisumasu?
• English Japanese
• kr?sm?s kurisumasu ?

7
Grammar

• Acoustic Form
• perception
  (e.g.,Boersma 1998, V Pater 2004)
• Phonological Representation gt UR gt
• production
• V
• Phonetic Representation

8
(1) Modification Misproduction (e.g., Paradis
and LaCharité 1997)

• Adapters correctly identify FL phonemes, map to
  UR.
• Production grammar repairs underlying
  representations to conform to native language
  constraints.

9
Problems with claim that all modification
misproduction

• Modification may be influenced by subphonemic
  information.
• e.g., Kang 2003 Vowels in Korean loans may be
  inserted even after legal stop codas. Likelihood
  of insertion is related to likelihood of the
  release of that stop in the English source.

10
(2) ModificationMisperception (e.g., Peperkamp
Dupoux 2003)

• Japanese listeners hear ebzo as ebuzo.
• Cf. Dupoux et al. 1999, Dehaene-Lambertz et al.
  2000, Jacquemot et al. 2003 for experimental
  support.

11
Not a lexical effect Dupoux et al. 2001

• Lexical decision task
• nonword stimuli real words
• sokdo sokudo speed
• mikdo mikado emperor
• Sokdo classified as real words, mikdo as nonwords

12
Problems with claim that all adaptation
misperception

• Some perception is accurate (e.g. Berent et al.
  syllable-counting experiments)
• Still must explain direction of misperception
  why kurisumasu and not kilihimete, etc.?

13
(3) Modification misproduction, but guided by
phonetic similarity

• P-map Hypothesis (Steriade 2001, etc.) Learners
  perceive FL phonological forms accurately, but
  the production grammar contains constraints that
  enforce phonetic similarity between UR and PR.

14
Problem with claim that adaptation production
perceptual similarity

• Even after adding perceptual similarity
  constraints to the production grammar, we are
  left with a residue of cases that must be
  analyzed as misperception (inaccurate mapping
  from foreign acoustic form to adapters UR).

15
(4) Dual-level model(e.g., Silverman 1992, Yip
2002, 2006)

• Listeners misperceive less salient features
  (partially inaccurate mapping to UR).
• Listeners accurately perceive more salient
  features, but production grammar may still make
  changes in mapping from UR to PR.

16
Problem with Dual-Level Model of adaptation

• Lack of clear criteria for deciding whether a
  particular modification pattern is a function of
• Misperception
• Misproduction

17
Questions, continued

• If foreign forms are misperceived, at what level
  of processing does this misperception occur?

18
Questions, continued

• To what extent is perception determined by early
  language experience?
• Is there a neural commitment to L1 contrasts?

19
Questions, continued

• Can formal theories of grammar shed light on
  foreign language production patterns?

20
One Potential Criterion for perception vs.
production Learnability

• OT aims to define
• What is a possible grammar (set of ranked
  constraints).
• What is a learnable grammar (rankings can be
  derived from input data, using an error-driven
  algorithm).

21

• Modification patterns that cannot be described in
  terms of learnable production grammar rankings
  must be a function of
• Misperception, or
• Other factors (frequency, timeline of exposure to
  FL, etc.).

22
Today

• Models of Acquisition First Language and Foreign
  Languages

23
To build a phonological grammar, children must
learn

• What is linguistically significant in the target
  language (possible contrasts).
• What is legal in the target language (possible
  structures, phonotactics).
• Morphemes and allomorphs (alternations).

24
Stages of Acquisition (e.g., Hayes 2004)

• Birth to 6 months can distinguish all possible
  phoneme contrasts.
• 6-8 months begin to form sound categories
  (perceptual magnet effects).
• 8-10 months begin to form a lexicon begin to
  learn phoneme categories of ambient language.
• Older begin to learn morphological processes,
  alternations.

25
Perception vs. Production Common Assumptions

• Children generally perceive L1 accurately.
• Many of childrens simplifications of adult forms
  are due to misproduction rather than
  misperception.

26
Example One Argument for Accurate Perception

• Gnanadesikan (2004) Productions by G, 27-33
  months

27
Simplification of onset clusters

• G Adult
• a. s-stop
• gaj skaj sky
• b?w sp?l spill
• d? star star
• b. s-sonorant
• so sno snow
• sip slip sleep

28

• s-obstruent gt obstruent (sky gt gaj)
• s-sonorant gt s (snow gt so)
• Onset C of lowest sonority is maintained.

29
Clusters containing labial r or w

• G Adult
• a. pi tri tree
• b. b?k dr??k drink
• c. paj kraj cry
• d. bep grep grape
• e. pajt kwajt quite
• f. f?'D? sw?'D? sweater
• g. f?w sm?l smell

30

• pr, tr, kr, kw gt p (tree gt pi)
• br, dr, gr gt b (grape gt bep)
• sm, sw gt f (smell gt f?w
• Labial articulation is always maintained, though
  labial segment may disappear.

31
Gs perception bep grape

• Does G actually hear (e.g.) gr as b?

32
Replacement of initial unstressed syllable

• G Adult
• a. fitén? k?ntén?r container
• b. fig?'Di sp?g?'Di spaghetti
• c. fib?'k? r?b?'k? Rebecca
• d. fimáwo t?máro tomorrow

33
Accurate perception?

• Does G actually hear the material in the initial
  syllable replaced by fi?

34
Preference for obstruent onsets

• G Adult
• a. fikál? koál? koala
• b. fibún b?lún balloon
• c. fipís p?lís police
• d. fibó b?ló below

35

• When the syllable following fi begins with a
  high sonority onset (liquid, glide) or no onset,
  the word-initial onset is recruited
• balloon gt fibún

36
Gs Perception

• Although G replaces initial unstressed syllables
  with fi, she apparently does hear the segmental
  content (at least the onset) of these
  syllables--because features of that onset may
  appear elsewhere in the word.

37

• fib?j? g?r?l? gorilla
• It seems unlikely that G hears gr in gorilla
  as b, since G seems to hear that gorilla
  contains 3 syllables.

38
Gnanadesikans analysis

• Gs modifications result from a grammar that
  differs from the adult grammar.
• These modfications result from a preference for
  less marked surface structures.
• Each feature of Gs grammar is attested in some
  adult NL grammar.

39
OT is intended as a theory of

• typologydefines possible grammars.
• learnability--defines how a grammar is learned
  from exposure to a set of data.

40
Architecture of the theory

• Markedness constraints (or well-formedness
  constraints, structural constraints) define
  possible surface structures.
• e.g. NoCoda, NoComplexOnset

41

• Faithfulness constraints define possible mappings
  from lexical representations to surface
  representations.
• e.g. Dep (no insertion), Max (no deletion)

42
Ranking Tableau MgtgtF
/sno/ NoComplexOnset (Markedness) Max (no deletion) (Faithfulness)
a. sno !
gt b. so
43
Ranking Tableau FgtgtM
/sno/ Max (no deletion) (Faithfulness) NoComplexOnset (Markedness)
gt a. sno
b. so !
44
Rankings determine surface structures

• M gtgtF suppresses surface contrasts.
• NoComplexOnset gtgt Max
• /so/ gt so
• /sno/ gt so
• (no CV-CCV contrast is possible)

45

• F gtgt M preserves lexical contrasts.
• Max gtgt NoComplexOnset
• /so/ gt so
• /sno/ gt sno
• (CV-CCV contrast is possible)

46
Rankings determine preferred repair

• Dep gtgt Max /sno/ gt so
• (deletion is preferred to insertion)
• Max gtgt Dep /sno/ gt sVno
• (insertion is preferred to deletion)

47
Rankings determine

• Choice of deletion vs. insertion
• Which C is deleted (/sno/ gt so vs. no)
• Which V is inserted (/sno/ gt sino, sono, etc.)

48
Assumptions (classical OT)

• Constraint set is universal.
• Rankings are language-specific.
• Each possible ranking defines a possible grammar.

49
Classical OT model of First Language Acquisition

• Constraints are innate.
• Rankings must be learned.

50
Corollary

• Each developing grammar (each stage of language
  acquisition) must represent a possible human
  grammar, since grammars differ only in ranking of
  constraints.

51
Parallels between Gs grammar and adult grammars

• onset simplification to lowest-sonority C gay
  for sky, so for snow
• Sanskrit reduplication pa-prach, tha-stha

52
Coalescence of segments

• bep for 'grape', f?w for 'smell
• Navajo dx gt g, Luganda mu gt mw, Kirundi
  tu gt tkw

53
Preference for trochaic feet, aligned with left
edge fibún for balloon

• Fikkert 1994, Demuth 1996
• Dutch óxant (ólifant), ánd?, ánR? (andere),
• bál? (bal)
• Sesotho kolo (sekolo) school
• Kiche (word-final stress) lóm (jolóm) head

54
Melodic overwriting fi

• Kolami pal-gil, kota-gita, maasur-giisur
• Chinese secret language may ka for ma, xway kwey
  for xwey

55
Child grammar possible adult grammar

• Each developing grammar should reflect some
  possible constraint ranking.

56
Question

• Is there an initial state/default constraint
  ranking?
• Answer from Gnanadesikan and others MgtgtF

57
Arguments for Default MgtgtF

• 1. Childrens modifications are generally in the
  direction of reduced markedness.

58
Subset Problem (Angluin 1980, Baker 1979)

• 2. If children can only use positive evidence
  (actual linguistic forms) in constructing a
  grammar, they must begin with the most
  restrictive grammar possibleotherwise their
  grammars will overgenerate.

59
Illustration

• Child C (for conservative) assumes MgtgtF
  NoComplexOnset gtgt Faithfulness.
• Child Cs grammar allows only CV syllables.

60

• Child R (for reckless) assumes FgtgtM
• Faithfulness gtgt NoComplexOnset.
• Child Rs grammar allows both CV and CCV
  syllables.

61
If Child C (MgtgtF) is born to

• Hawaiian-speaking parents,
• Childs grammar adult grammar
• English-speaking parents,
• Childs grammar ? adult grammar, but no
  worries--child gets positive evidence (CCV)
  telling her to rerank constraints.

62
If Child R (FgtgtM) is born to

• English-speaking parents,
• Childs grammar adult grammar
• Hawaiian-speaking parents,
• ? Childs grammar ? adult grammar, AND
• no positive evidence can ever trigger
• reranking.

63
Learning is error-driven

• Default ranking MgtgtF.
• Other rankings (MgtgtM, FgtgtF) must be learned from
  data.

64
What is default for language contact situations?

• Presumably, the learner/adapter begins from the
  NL rankings.
• Therefore, adaptation/error patterns should be
  explainable as either
• transfer of NL rankings, or
• universal default rankings.

65
Potential Problems in Language Contact Phonology

• 1. M gtgtM rankings (differential difficulty)
• NL bans 2 structures
• FL allows both structures

66

• BUT learners/adapters master one structure more
  easily.
• e.g. Japanese shiitibanku (both ti and si are
  illegal in Japanese).

67
2. FgtgtF rankings (differential repair)

• NL has no inputs with illegal structures, so no
  evidence for repair preference
• BUT learners/adapters adopt specific repairs,
• AND these repairs may vary across languages
  (therefore not universal),
• AND different repairs may be used within a single
  language in different contexts.

68
Dehu (Tryon 1970)

• a. Obstruent__Sonorant copy V
• peleit plate
• galas glas
• b. Obstruent__Obstruent default i
• sipö spur
• sipun spoon

69
3. Ranking Reversals

• Korean NL stopnasal gt nasal nasal
• /kukmul/ gt ku?mul soup
• But in SLA, Koreans often insert vowel
• /tegnal/ gt teg?nal (Hwang 2006).

70
Proposal

• Where modification patterns would require a
  grammar with unlearnable rankings, these patterns
  have their source in factors such as
• misperception
• frequency
• orthography
• time course of language contact
• etc.