Lecture3-SpeechPerception1

Lecture3-SpeechPerception1 - Psych 215L: Language...

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Psych 215L: Language Acquisition Lecture 3 Speech Perception Sounds of Language (Speech Perception) Learner’s job: Identify phonemes (contrastive sounds that signal a change in meaning) Phonemes are language-specific - r/l is a phonemic contrast in English but not in Japanese Kids of the world require knowledge of phonemes before they can figure out what different words are - and when different meanings are signaled by different words big vs. pig Lisa = Risa for some of my Japanese friends Learning Sounds About Speech Perception Important: Not all languages use the same contrastive sounds. Languages draw from a common set of sounds (which can be represented by the International Phonetic Alphabet (IPA)), but only use a subset of that common set. Child’s task: Figure out what sounds their native language uses contrastively. meaningful sounds in the language: “contrastive sounds” or phonemic contrasts Phonemic Constructed Acoustic Innate Speech Perception: Computational Problem Speech Perception: Computational Problem Real world data are actually much harder than this… (from Swingley 2009) Divide sounds into contrastive categories (phonemes) Here, 23 acoustically-different sounds are clustered into 4 contrastive categories. Sounds within categories are perceived as being identical to each other. x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x C2 x C1 x x x C4 x x x x x C3 x x x x Categorical Perception Categorical perception occurs when a range of stimuli that differ continuously are perceived as belonging to only a few categories with no degrees of difference within a given category. Actual stimuli Categorical Perception of stimuli Acoustic-Level Information Includes: timing and frequency Tones: frequency (close-up) Acoustic-Level Information Language sounds Acoustic-Level Information Language sounds Male Vowels (close up) Vowels combine acoustic energy at a number of different frequencies Different vowels ([a] “ah”, [i] “ee”, [u] “oo” etc.) contain acoustic energy at different frequencies Listeners must perform a frequency analysis of vowels in order to identify them (Fourier Analysis) Acoustic-Level Information Language sounds Female Vowels (close up) Synthesized Speech Allows for precise control of sounds Valuable tool for investigating perception Acoustic-Level Information Language sounds Timing: Voicing Acoustic-Level Information Language sounds Timing: Voice Onset Time (VOT) 60 ms English VOT production Dutch Not uniform - there are 2 categories (distribution is bimodal) Eastern Armenian Spanish Thai Hungarian Korean Tamil Cantonese Perception of stimuli: 2 categories Hindi English Marathi Perceiving VOT Discrimination Task “Are these two sounds the same or different?” ‘Categorical Perception’: dQ vs. tQ More uncertainty/ error at category boundary Longer decision time at category boundary Same/Different 0ms 60ms Same/Different 0ms 10ms Decision between d/t Time to make decision Identification task:“Is this sound dQ or tQ?” Discrimination Task “Are these two sounds the same or different?” Same/Different 40ms 40ms Discrimination Task “Are these two sounds the same or different?” D Same/Different 0ms 10ms Why is this pair difficult? 0ms 20ms D D Same/Different 0ms 60ms 20ms 40ms T T 40ms 60ms T (i) Acoustically similar? Same/Different 40ms 40ms Across-Category Discrimination is Easy (ii) Same Category? Within-Category Discrimination is Hard Cross-language Differences Cross-Language Differences Miyawaki et al. 1975 Identification task: R English speakers can discriminate r and l, and seem to show a similar pattern of categorical perception to what we saw for d vs. t L R -----------------------> L L R Cross-Language Differences Discrimination task: English speakers have higher performance at the r/l category boundary, where one sound is perceived as r and one sound is perceived as l. Japanese speakers generally perform poorly (at chance), no matter what sounds are compared because r and l are not contrastive for them. Cross-Language Differences Hindi dental [d] (tip of tongue touches back of teeth) ? Miyawaki et al. 1975 retroflex [D] (tongue curled so tip is behind alveolar ridge) English [d] is usually somewhere between these Infant Speech Perception Cross-Language Differences Salish (Native North American language): glotalized voiceless stops Uvular – tongue is raised against the velum How do we tell what infants know, or use, or are sensitive to? Researchers use indirect measurement techniques. Some information from the High Amplitude Sucking (HAS) paradigm (Eimas et al. 1971) Infants have sophisticated discrimination abilities, but they don’t abstract sounds into categories the way that adults do. Velar – tongue is raised behind the velum (they are actually ejectives - ejective is produced by obstructing the airflow by raising the back of the tongue against or behind the velum) Infant Speech Perception How do we tell what infants know, or use, or are sensitive to? Researchers use indirect measurement techniques. Adult perception “tQ” phonemic category How do we tell what infants know, or use, or are sensitive to? Researchers use indirect measurement techniques. Some information from the High Amplitude Sucking (HAS) paradigm (Eimas et al. 1971) Infants have sophisticated discrimination abilities, but they don’t abstract sounds into categories the way that adults do. “dQ” Infant Speech Perception phonemic category Some information from the High Amplitude Sucking (HAS) paradigm (Eimas et al. 1971) Infants have sophisticated discrimination abilities, but they don’t abstract sounds into categories the way that adults do. Infant perception “dQ 1” “dQ 2” “tQ 1” “tQ 2” Perceiving sound contrasts Speech Perception of Non-Native Sounds Comparing perceptual ability Kids… Kids… Werker et al. 1981: English-learning 6-8 month olds compared against English & Hindi adults on English & Hindi contrasts This ability to distinguish sound contrasts extends to phonemic contrasts that are nonnative. (Japanese infants can discriminate contrasts used in English but not in Japanese, like r/l.) This goes for both vowels and consonants. Conditioned Head Turn Procedure …vs. adults Adults can’t, especially without training - even if the different is quite acoustically salient. So when is this ability lost? And what changes from childhood to adulthood? Werker (1995): Speech Perception But when after 6-8 months is the ability to lost? Werker & Tees (1984) Key into “critical period” hypothesis for language (Lenneberg 1967) - when language can be learned natively “To test for this critial period, children of 12 and 8 years were tested, with the expectation that the 8year-olds but not the 12-year-olds would be able to discriminate nonnative contrasts. English-speaking children of both ages, however, performed like English-speaking adults…study was extended to 4year old children, who actually performed most poorly of all on nonnative contrasts….findings revealed that experience must begin to influence speech perception long before 4, certainly well before the critical period suggested by Lenneberg.” Speech Perception of Non-Native Sounds But when after 6-8 months is the ability to lost? Salish & Hindi contrasts Change happens somewhere around 8-10 months, depending on the sound contrast. See Yoshida et al. (2010) for evidence that infants have some malleability still at 10 months, but it’s much less than at 6 or 8 months. Werker & Tees (1984) Discovering contrastive sounds: What’s the point of it again? The idea is that once children discover the meaningful sounds in their language, they can begin to figure out what the words are. Ex: An English child will know that “cat” and “caat” are the same word (and should have the same meaning). As adults, we can look at a language and figure out what the contrastive sounds are by looking at what changes a word’s meaning. But children can’t do this - they figure out the contrastive sounds before they figure out words and word meanings. Experimental Study: Dietrich, Swingley & Werker (2007) Testing children’s perception of contrastive sounds Dutch and English contrastive features differ. In English, the length of the vowel is not contrastive “cat” = “caat” In Dutch, the length of the vowel is contrastive More about contrastive sounds There are a number of acoustically salient features for sounds. All it takes for sounds to be contrastive is for them to have “opposite” values for one feature. Example: English sounds “k” and “g” differ only with respect to voicing. They are pretty much identical on all other features. Many contrastive sounds in English use the voicing feature as the relevant feature of contrast (p/b, t/d, s/z, etc.). However, there are other features that are used as well (air flow, manner of articulation, etc.). Task for the child: Figure out which features are used contrastively by the language. Contrastive sounds for the language will usually vary with respect to one of those features. Does the data distribution show this? Dutch and English vowel sounds in the native language environment also seem to differ “…studies suggest that differences between the long and short vowels of Dutch are larger than any analogous differences for English.” Frequency of sound in input Dutch English “cat” ≠ “caat” (Japanese also uses this feature) 0 Vowel duration Does the data distribution show this? Does the data distribution show this? Dutch and English vowel sounds in the native language environment also seem to differ Dutch and English vowel sounds in the native language environment also seem to differ “…studies suggest that differences between the long and short vowels of Dutch are larger than any analogous differences for English.” “…studies suggest that differences between the long and short vowels of Dutch are larger than any analogous differences for English.” Dutch vowel length used contrastively; vowels tend to be either very short or very long Frequency of sound in input 0 Dutch English English vowel length not used contrastively; vowels tend to be less short and less long (comparatively) Frequency of sound in input 0 Vowel duration Does the data distribution show this? Dutch English Vowel duration Does the data distribution show this? Dutch and English vowel sounds in the native language environment also seem to differ Dutch and English vowel sounds in the native language environment also seem to differ “…studies suggest that differences between the long and short vowels of Dutch are larger than any analogous differences for English.” “…studies suggest that differences between the long and short vowels of Dutch are larger than any analogous differences for English.” Dutch = bimodal distribution? English = unimodal distribution? Frequency of sound in input 0 Vowel duration Dutch English Dutch = bimodal distribution? English = unimodal distribution? Frequency of sound in input 0 Vowel duration Dutch English Learning from real data distributions Maye, Werker, & Gerken (2002) How do we know that children are sensitive to distributional information? Created synthetic sounds ranging from [da] to [ta] that were nonnative for the infants (because they were unaspirated). Maye, Werker, & Gerken (2002) • Familiarized 6 to 8-month-old infants to one of two sets – Bimodal Set: Sounds on the ends near [da] and [ta]. – Unimodal Set: Sounds in the middle. • Test preference for: – 3 6 3 6… (Alternating) vs. 3 3 3 3… (Non-alternating) stimuli Maye, Werker, & Gerken (2002) 3333 3636… = = < < Maye, Werker, & Gerken (2002) Maye, Werker, & Gerken (2002) Infants trained on the Bimodal data had a novelty preference for non-alternating trials. They learned to expect alteration, and were surprised by nonalteration. Infants trained on the Unimodal data did not prefer/disprefer one over the other. The did not seem to learn any expectation. 3333 3636… = = < < = = < < Back to Dietrich, Swingley, & Werker (2007) Dutch and English vowel sounds in the native language environment also seem to differ “…studies suggest that differences between the long and short vowels of Dutch are larger than any analogous differences for English.” Dutch = bimodal distribution? English = unimodal distribution? Frequency of sound in input 3333 3636… Back to Dietrich, Swingley, & Werker (2007) Prediction if children are sensitive to this distribution Dutch children interpret vowel duration as a meaningful contrast because the distribution is more bimodal Implication: Change to vowel duration = new word Dutch English English children should not interpret vowel duration as a meaningful contrast because the distribution is more unimodal Implication: Change to vowel duration = same word as before 0 Vowel duration Dietrich, Swingley, & Werker (2007) Tests with 18-month-old children who know some words (and so have figured out the meaningful sounds in their language) Dietrich, Swingley, & Werker (2007) Experiment 1: Testing English and Dutch kids on Dutch vowel durations Frequency of sound in input 0 “Switch” Procedure: measures looking time …this is a tam…look at the tam Vowel duration Dutch kids 5.04 sec 9.23 sec Habituation difference English kids 6.66 sec 7.15 sec Same: look at the tam! Switch: look at the taam! Test Same: look at the tam! no difference Switch: look at the taam! Test Dietrich, Swingley, & Werker (2007) Dietrich, Swingley, & Werker (2007) Experiment 2: Testing English and Dutch kids on English vowel durations Experiment 3: Testing English and Dutch kids on vowel quality contrast (a/e) Frequency of sound in input Frequency of sound in input 0 0 Vowel duration Dutch kids 5.92 sec 8.16 sec English kids 7.34 sec 8.04 sec Same: look at the tam! Test difference no difference (This is a control condition to make sure English kids can do the task when the sound is contrastive for them) Switch: look at the taam! Vowel duration Dutch kids 4.08 sec 5.72 sec English kids 6.31 sec 9.31 sec Same: look at the tam! Test difference Switch: look at the tem! difference Dietrich, Swingley, & Werker (2007) Just a note that experimental data with infants is messier than it sounds. Dietrich, Swingley, & Werker (2007) Implications of experiments 1, 2, and 3: Dutch children recognize vowel duration as contrastive for their language while English children do not. This can only be due to the data encountered by each set of children in their language. Dutch children have a category boundary approximately here. English children do not. Frequency of sound in input Dutch English 0 What drives children to learn the distinction? “One frequently raised hypothesis…is that it is driven by contrast in the vocabulary. Dutch children might learn that [a] and [a:] are different because the words [stat]…and [sta:t]…mean different things…however, children that young do not seem to know many word pairs that could clearly indicate a distinction between [a] and [a:].” Vowel duration Dietrich, Swingley, & Werker (2007) “The other current hypothesis is that children begin to induce phonological categories “bottom-up”, based on their discovery of clusters of speech sounds in phonetic space…undoubtedly implicated in infants’ early phonetic category learning, which begins before infants know enough words for vocabulary-based hypotheses to be feasible…” Dietrich, Swingley, & Werker (2007) Swingley (2009) “A necessary condition for such learning to be the driving force behind Dutch children’s phonological interpretation in the present studies is that long and short vowels be more clearly separable in Dutch than in English…preliminary examination of this problem using corpora of Dutch child-directed speech indicated that the set of long and short instances formed largely overlapping distributions.” One potential source of information: keep some contextual information for each vowel sound (what word it came from, if it comes from a frequent word). Frequency of sound in input Dutch English 0 Vowel duration Implication: Dutch children need other cues to help them out ...
View Full Document

This note was uploaded on 12/12/2011 for the course PSYCH 215l taught by Professor Pearl during the Fall '11 term at UC Irvine.

Ask a homework question - tutors are online