Presentation is loading. Please wait.

Presentation is loading. Please wait.

1 Pragmatic & Perceptual Biases on Phoneme Identification Young Ah Do (MIT Linguistics) TedLab. BCS. MIT 25 th April 2012.

Similar presentations


Presentation on theme: "1 Pragmatic & Perceptual Biases on Phoneme Identification Young Ah Do (MIT Linguistics) TedLab. BCS. MIT 25 th April 2012."— Presentation transcript:

1 1 Pragmatic & Perceptual Biases on Phoneme Identification Young Ah Do (MIT Linguistics) TedLab. BCS. MIT 25 th April 2012

2 Introduction  This study is about the integration of multiple linguistic sources Effects of; Pragmatic Bias on Phoneme Identification Pragmatic Bias & Phonetic Bias on Phoneme Identification 2

3 Integration of Linguistic Knowledge  In sentence and speech processing, multiple sources of linguistic knowledge across multiple levels of linguistic structure integrate. (examples follow). 3

4 Sentence Processing  Syntactic cues only  Fodor, Bever & Garrett (1974), Frazzier & Fodor (1978)  Semantics cues  Grodner, Gibson & Watson (2005)  Lexical information  Fedrenko, Piantadosi & Gibson (in press) 4

5 Speech Processing  Phonotactic knowledge to phoneme identification  Massaro and Cohen (1983), Pitt (1998) & Moreton (2002): phoneme identification towards phonetically legal sequences (Integration of low-level features)  Lexical knowledge to phonetic categorization  Ganong (1980): phonetic categorizations towards answers that would make a real word  Pitt (1995) 5

6 Language Processing  What we know; Multiple sources of linguistic knowledge across linguistic levels communicate and interact (Marslen-Wilson & Tyler 198) 6

7 Language Processing  What we don’t know yet;  Q1: The maximum range of integration.  Q2: How do comprehenders use multiple sources in a processing task? 7

8 Introduction again  This study is about the integration of multiple linguistic sources Effects of; Pragmatic Bias on Phoneme Identification Pragmatic Bias & Phonetic Bias on Phoneme Identification 8

9 The maximum range of integration  How far can the integration of the distinctive linguistic levels expand in language processing? Can Pragmatics affect Phonetics ?  Assumptions;  Phonetics and Pragmatics are at the two far ends of the linguistic structure. 9

10 Multiple sources  Previous studies observed integrated effects of one linguistic source to a specific language processing  If multiple biases are given simultaneously, how would comprehends use multiple cues? Will the cues in different levels interact?  Effects of Pragmatic bias & Phonetic bias on Phoneme Identification 10

11 A Previous Finding on Q1  Rohde & Ettlinger (R&E, 2010)  Experimental investigation with English speakers  Task: pronoun interpretation  Phonetics : words on the [hi]~[∫i] continuum.  Pragmatics : expectation motivated by discourse context 11

12 A Previous Finding on Q1  Sally helped Bruce because ? was working on the same project (R&E 2010, p.384).  Pragmatically, [he] biasing context  Sally annoyed Bruce because ? was in a bad mood (R&E 2010, p.384).  Pragmatically, [she] biasing context 12

13 Pragmatics to Phonetics  Results: pronoun interpretation is guided by pragmatic bias.  Pragmatic inference can alter listeners’ identification of phonetic categories. 13

14 This study  Effects of pragmatic inference on phoneme identification.  Pragmatics: methods replicated from R&E.  With adding phonetic bias as well.  Korean 14

15 This study  A Case study in Korean  Pragmatic cues- implicit causality (IC) verbs  Perceptual bias in different conditions – velars before front and back vowels 15

16 Outline of the Talk  Background on Implicit Causality verbs  Experiment 1 : Sentence completing task with an Implicit causality verb  Background on Velar Palatalization  Experiment 2 : Perception of velar stops and the corresponding palatoalveolar affricates before various vowels  Experiment 3 : pronoun interpretation using pragmatic (Exp. 1) and phonetic (Exp. 2) biases.  Integration of multiple linguistic sources 16

17 Implicit Causality (IC) Verbs  IC verbs (Garvey & Caramazza 1974)  ‘guide listeners’ coreference expectations by describing events in which one participant (either the subject or object, depending on the verb) is implicated as central to the event’s cause and is thus likely to be re-mentioned in a subsequent because clause’ (emphasis is mine). 17

18 Implicit Causality (IC) Verbs  Sally helped Bruce because …  Object-biasing context  Sally annoyed Bruce because …  Subject-biasing context 18

19 Experiment 1  Test whether Korean listeners’ coreference expectations are affected by IC verbs. 19

20 Methods  Participants  29 adult Korean  Materials  36 sentences consisting of two clauses connected with because.  The 1 st clause introduces two individuals with opposite gender and an IC verb.  18 sentences introduce subject-biasing, 18 for object- biasing IC verbs. 20

21 Methods  Materials  Hyesun-i Daeyoung- ɨ l towa-t-ta, wenjahamy ə n ____- i __________. Hyesun- NOM Daeyoung- ACC help- PAST - DECL, because ____- NOM ________________. Hyesun(female) helped Daeyoung(male), because ____________.  Procedure  Participants were asked to complete the 2 nd clause by typing, after reading the 1 st clause on the screen.  One sentence was given at a time in random order. 21

22 Data Analysis  Analysis of production data by  Repetition of the noun: Hyesun  Gender matching pronoun: She  Predicate when a noun is unclear : has nothing else to do 22

23 Results  Sentences with subject-biasing contexts yielded higher subject repeating rating than object-biasing contexts. 23

24 Findings  Korean comprehenders do expect different coreferential nouns in the following clause introduced by because, depending on IC verbs in the preceding clause.  Pragmatic context (IC verbs) has effect on speakers’ expectation in sentence processing. 24

25 Going back to the question  Effects of pragmatics on phoneme identification  Findings from Experiment 1;  Validity of pragmatic bias (IC verbs) 25

26 Going back to the question  Effects of phonetic bias on phoneme identification  Phonetic bias created by perceptual confusion 26

27 Perceptual confusion  Stimuli again  X helped Y, because ? (either X or Y )________.  Design  X and Y are perceptually confusable with various degrees 27

28 Velar Palatalization  Velar palatalization is the change from a velar stop consonant, voiceless [k] or voiced [g], to the corresponding palatoalveolar affricate, voiceless [t ʃ ] or voiced [d ʒ ].  electri[k] vs. electri[s]ity  publi[k] vs. publi[s]ity 28

29 Velar Palatalization  A velar stop consonant is more confusable with its corresponding palatoalveolar affricate before a front vowel (Wilson 2006).  Δ (ki ~ t ʃ i) < Δ (ka ~ t ʃ a), Δ (ko ~ t ʃ o), Δ (ku ~ t ʃ u)  Δ (gi ~ d ʒ i) < Δ (ga~ d ʒ a), Δ (go ~ d ʒ o), Δ (gu ~ d ʒ u) 29

30 Velar Palatalization  A velar stop consonant and the corresponding palatoalveolar affricate are similar more before front vowels, both acoustically (Keating and Lahiri 1993) and perceptually (Guion 1996, 1998).  Confusions observed from an experimental study in English (Guion 1998) 30

31 Confusion matrix (Guion, 1998. p.55) 31

32 Modeling perceptual bias  Korean shows pairs of last names whose initial consonant is velar stop and palatoalveolar affricate before various vowels.  Ki, Chi  Ko, Cho  Ku, Chu  Kang, Chang  All of them are among 40 most frequent Korean last names. 32

33 Velar Palatalization in Korean  Different from English, Korean lexicon does not show morphologically or lexically conditioned Velar Palatalization.  Korean speakers do not have lexical evidence of velar palatalization. 33

34 Velar Palatalization in Korean  Will speakers who do not have lexical evidence of velar palatalization be biased to perceive [k] as [ch] more before a front vowel?  Yes, if it is substantive bias (Wilson 2006).  No, if it is lexically-motivated bias. 34

35 Experiment 2  Participants  Subset of participants in Experiment 1(23 among 29)  Materials  16 pairs of nonce words whose initial consonants are velar stop and palatoalveolar affricates.  In a sentence such as ‘It is _____’. 35

36 Methods  kal/chal ibmi-da. kal/chal be-decl. (It) is a [kal]/[chal].  ki/chi ibmi-da. ki/chi be-decl. (It) is a [ki]/[chi]. 36

37 Methods  Materials  16 nonce word pairs were one syllable either of CV or CVC structure and V are [i], [a], [o] and [u].  2 Korean male speakers read sentences.  Procedure  Participants listened sentences in random order.  They were asked to decide whether the word they heard is [K] initial, [Ch] initial or they cannot be sure. 37

38 Confusion matrix  More confusion before [i] than before other vowels. 38 KiChiKaChaKoChoKuChu Ki4832 Chi693 Ka 797 Cha 1082 Ko 6328 Cho 1678 Ku 7216 Chu 1183

39 Findings  Even without lexical evidence of velar palatalization, Korean speakers show greater confusion between [k] and [ch] before a front vowel than before other vowels.  Velar stops and palatoalveolar affricates before various vowels are valid pair to use in designing perceptual confusion. 39

40 Interim Summary  For Korean listeners,  IC verbs are valid for designing pragmatic bias: Experiment 1.  [k] and [ch] in various conditions is valid for designing perceptual bias: Experiment 2. 40

41 The questions  Do listeners use pragmatic cues in identifying phonemes ?  How do listeners use pragmatic and phonetic cues if they are imposed simultaneously in sentence processing?  Will pragmatic and phonetic cues interact?  Experiment 3 is to answer this question. 41

42 Experiment 3  Participants  Same as in Experiment 2 (23 Korean native speakers)  Exp 3 was done before Exp 2 and Exp 1.  Materials  32 sentences consisting of two clauses connected by because.  1 st clauses presented in Experiment 1 were reused in constructing 1 st clauses in Experiment 3. 42

43 Methods  Materials  The 1 st clause introduces 2 individuals with [k] and [ch] initial names and an IC verb.  18 sentences introduce subject-biasing, 18 for object- biasing IC verbs.  In the 2 nd clause after because, either [k] initial or [ch] initial pronoun that were presented in the proceeding clause was given. 43

44 Methods  Materials  In the 2 nd clause after because, either [kV] or [chV] pronoun that was presented in the proceeding clause was given.  In order to avoid perceptual confusion in the 1 st clause, sentences were shown on the screen.  2 male speakers read sentences. 44

45 An Example  Materials Ki -ssi-ka Chi -ssi-l ɨ l hwanagehe-t-ta, wenjahamy ə n Ki/Chi -ssi-ka …. Ki-Mr- NOM Chi-Mr- ACC annoy- PAST - DECL because Mr-Ki/Chi- NOM …. Mr. Ki annoyed Mr. Chi, because Mr. Ki/Chi …. 45

46 Pragmatic bias & Phonetic bias 46

47 Procedure  Procedure  Participants listened a sentence shown on the screen and decided what they think is produced in [kV]/[chV] position.  A forced choice testing between [k] and [ch]. 47

48 Pragmatics, Perception & Phonetics 48 Phonetics

49 Pragmatics, Perception & Phonetics 49 Phonetics

50 Statistic analysis  A mixed logit model for phoneme identification ([k] vs. [ch])  Fixed factors  Pragmatic Bias (subj-bias vs. obj-bias)  Phonetic Bias (vowel backness)  Audio (heard item)  Name frequency  Phonotactic frequency  Perceptual experiment bias (experiment 2) 50

51 Findings  A mixed logit model fitted with contrast coded fixed effects for phoneme identification ([k] vs. [ch])  random factors  Subjects  Items 51

52 Results Factors Estimate S.D t value p value (Intercept) 0.01124 0.29238 0.038 0.969344 PragmaticBias 3.08509 0.57102 5.403 6.56e-08 *** Audio 5.41820 0.57111 9.487 < 2e-16 *** PerceptualBias 1.86958 0.51988 3.596 0.000323 *** PragBias:Audio -0.24915 1.14202 -0.218 0.827297 PragBias:PercepBias 1.35651 1.03974 1.305 0.192009 Audio:PerceptBias 2.67674 1.03975 2.574 0.010041 * PragBias:PerceptBias:Audio-1.14178 2.07951 -0.549 0.582962 52

53 Results Pragmatic bias p <.0001 Audio (which sound was played) p <. 0001 Phonetic bias (vowel environment) p <.0001 Prag:Audio p >.5 Prag:Phonetics p >.1 Audio:Vowel p >.01 Prag:Audio:Phonetics p >.5  No significant model improvement with adding the following factors;  Name frequency (p =.102)  Phonotactic bias (p=.128)  Perceptual experiment bias (p=.067) 53

54 Discussion  In sentence processing, listeners’ expectation are affect by IC verbs.  Velar stop and its corresponding palatoalveolar affricate is more confusable before high front vowels than before other vowels, even among speakers whose language does not show velar palatalization; velar palatalization as universal pattern. 54

55 Discussion  Pragmatic bias influences phoneme identification.  Perceptual bias influences phoneme identification.  Listeners use both biases simultaneously in phoneme identification task.  But pragmatic bias and phonetic bias do not interact super-additively. 55

56 Thanks! youngah@mit.edu 56


Download ppt "1 Pragmatic & Perceptual Biases on Phoneme Identification Young Ah Do (MIT Linguistics) TedLab. BCS. MIT 25 th April 2012."

Similar presentations


Ads by Google