Speech perception Flashcards
When a fetus can perceive environmental sounds
7 months of gestation
Fetuses only perceives ___low/high frequencies
Low (below 400Hz)
Higher frequencies speech (above 1000Hz) contains the most …
Acoustic information about segments (phones)
Fetuses receive more ____ information
Prosadic
Prosodic cues
- Pauses
- hanges in pitch (fundamental frequency (F0)
- Lengthening of segments or syllables
etc.
Infants as old as _____ are sensitive prosodic cues of word/phrase/clause boundaries.
3 days old
Learning from the environment speech perception hypothesis
Infants are not born with speech perceptual abilities
Predictions :
- Gradual development of the ability to distinguish speech sounds, and can only discriminate sounds in the native language (meaning there IS crosslinguistic differences)
Learning based on innate abilities speech perception hypothesis
Infants are born with speech perceptual abilities
Predictions :
- Infants are born with the ability to distinguish all speech sounds across languages. This ability reduces as the child grows.
Correct speech perception hypothesis
Between hypothesis of innateness and environment learning
- Infants are born with the ability to distinguish many non-native sounds but not all, some abilities are gained through perception of the input
What happens when a consonant is released
The articulators restricting airflow come apart.
Aspiration
Brief moment in which the vowel is not voiced after a voiceless stop
Voice onset time (VOT)
Period of time between the release of the stop and the onset of voicing (silence)
A VOT of 25 ms means…
Voicing begins 25 seconds after the stop is released
A VOT of more than ___ ms yields aspiration in English
25
A VOT of 0 means …
Voicing begins simultaneously with the release of the stop, creating a plain consonant (neither aspirated nor voiced)
A negative VOT means…
The voicing begins before the stop is released : the consonant AND the vowel are voiced
VOT boundary between voiced /b/ and voiceless /p/ in English
+25 ms
3 conditions of the VOT experiment on infants
1a. /ba/ vs /ba/ (VOT of -20 vs 0); voiced
1b. /pa/ vs /pa/ (VOT of +60 vs +80); aspirated (french/spanish boundary)
2. /ba/ vs /pa/ (VOT +20 vs +40); different
Steps of the VOT experiment
- First sound
- Habituation (HAS decreases)
- Second sound
- If the second sound is perceived as different, suck rate would increase (dehabituation)
Results of the VOT experiment
Even if all the pair of sounds only differed from 20ms, the infants heard only the difference between different categories present in English adults’ grammar (/b/ vs /p/)
Categorical perception
Infants hear sounds categorically
True or false ? Infants with a native language with a different VOT value as English could perceive the English voiced/voiceless categories
True
True or false ? Infants with English as their native language could distinguish Spanish voiced vs plain sounds
False
American infants may distinguish Spanish voiced vs plain sounds if the acoustic difference is ___
Larger (70-80 ms)
Explanation for why English exposed infants cannot discriminate the Spanish voiced/plain sounds
- Maybe Spanish boundary needs to be learned because it is marked (still 40-50% of languages have the same boundary)
- Maybe younger infants CAN discriminate the Spanish sounds, but lose the ability when these sounds are not in the target language
Spanish VOT boundary
-5 ms (contrast voiced vs voiceless is more defined), harder to produce but not perceive
Until __ months, babies perceive non-native phonemic contrasts almost perfectly
6
Perceptual narrowing (drop in nonnative contrasts perception performance)
10 months old : 50% lost
1 year old : below 20% : almost as poor as adults, coincides with development of the grammar of the native language
Age at which a baby’s ability to discriminate sounds is limited to the native language
1 year old
Maintenance-loss view
Idea that the neural structures used in the brain to respond to nonnative sound contrasts become inactive
Perceptual assimilation
Idea that our perceptual system is reorganized so that sounds that are not phonetically similar to native sounds are not even heard as linguistic sounds, while sounds that are similar to native sounds are assimilated to familiar sounds.
3 components of prosody
pitch, rythm, pauses
Babies of __ months know that pauses occur between rather than within clauses (subject+predicate)
7
Babies at __ months prefer IDS with pauses between rather than within phrases
9
Babies use ___ to find units in language
Pauses
From __ to 9 months, babies begin to break down units into smaller ones (from clause to phrase)
6
Prosodic/phonological bootstrapping
Babies notice patterns in the prosody of their language such as pause locations
____ create rhythm in language
Syllables
Syllables are defined as peaks of _____
Sonority
Nucleus of a syllable
Vowel
Rime
Nucleus + coda
A syllable with ____ stress usually indicates a word boundary
Primary
Speech segmentation
Identifying units in the continuous speech input
Age at which babies can segment speech into bisyllabic units with primary stress
7 to 9 months old (12 to 16 for French)
Phonotactics
Constraints on which speech sounds can occur next to each other in a syllable or a word
Transitional probabilities
Probability of 2 syllables to co-occur in a word (high transitional probability = syllables can probably occur together in a word)
9 months old infants prefer to listen to nonword sequences contaning…
Possible and highly likely phonotactic patterns
Babies listen longer to words presented in the midst of ____poor/good phonotactic cues
Good
Babies assume word boundaries where __low/high transitional probability is detected
Low (ignore places with high transitional probability)
In an experiment, babies heard strings of 4 made-up words, then were presented with words with different transitional probabilities :
E.g.
Whole word : golatu: go+la TP 1.0 (probability of 1 or 100% that go is preceded by la, and go is followed by la)
Part word : tudaro: tu+da TP 0.25 (between words : tu + the second word that could have been any of the 4 words, so 1/4 prob.)
How did the babies react to hearing part words?
They were surprised when they heard the part-words : they listened to them longer.
This suggests there is statistical learning : the infants were more interested in the part words because they were tracking co-occurances in the string of made up words
8 months old can identify plausible word units when there is high _____ _____ between syllables
Transitional probabilities
After hearing ABA words (ga ti ga) and ABB words (ga ti ti), and then hearing a different word of the same or opposite pattern, babies listened longer to the ____
different patterned word
Conclusion of the ABA/ABB study
Children can track the frequencies of abstract structures (even without using transitional probabilities)
True or false : infants can discriminate many but not all non-native contrasts
True
Infants gradually lose discrimination abilities for ______ contrasts.
Non-native
Czech and French contrast test
10 weeks old English exposed infants were tested on two contrasts: [pɑ]-[pɑ̃] oral vs nasal in Québec French and [ʒa]-[r̝a] in Czech
Result : They could distinguish between sounds in both pairs (even for the extremely marked Czech features)
Salish and Hindi contrasts tests
English and Salish exposed infants were tested on Salish contrasts
English and Hindi exposed infants were tested on Hindi contrasts
Results :
English-exposed infants: - Discrimination of both contrasts was high for 6-8mos ; decreased gradually at 8-10mos and at 10-12mos
* Salish- and Hindi-exposed infants: Accurate on their native contrasts
Infants are born with the ability to discriminate many human speech sounds, but discrimination of some contrasts requires…
Input from the native language (e.g. spanish voicing require exposure to Spanish)
[bi] vs [di] and [lɪf] vs. [nim] test
[bi] and [di] : associated with objects or not
[lɪf] vs. [nim] : associated with objects
Result :
- Children did not discriminate between [bi] and [di] when they were associated with objects because they were trying to learn the meaning of the words
- Children discriminated between [bi] and [di] when they were NOT associated with objects
- Children discriminated between [lɪf] vs. [nim] even when associated with objects because they were different enough (not minimal pair unlike [bi] and [di])
Conclusion of [bi] vs [di] and [lɪf] vs. [nim] test
- Infants can discriminate between 2 utterances even when they cannot associate a meaning with them.
- Phonological representations are reduced in segmental complexity compared to what infants perceive when trying to associate meaning to sounds : e.g.[b]+[ɪ] vs. [d]+[ɪ] can be stored as : [stop]+[ɪ] vs. [stop]+[ɪ]
Contrasts involving _____place/manner of articulation are generally acquired in perception earlier than contrasts involving _____place/manner of articulation
Manner of articulation : acquired quicker than place of articulation
2 possible reasons [h] is produced late
- Because of /h/ deletion rule
- [h] is hard to perceive
Some developmental changes in a child’s grammar system apply ‘across-the-board’, meaning…
They apply to all relevant words with variation only during a short period
When there is perceptual miscoding, developmental changes do not apply …
Across-the-board
True or false : production data can be used to inform perception understanding
True
Question that arises from Amahl’s production :
puzzle’ → [pʌdəɫ]
‘puddle’ → [pʌɡəɫ]
If Amahl can produce the [d] in puzzle, why doesn’t he produce it in puddle ?
Amahl’s production of puddle comes from a perception problem, meaning…
[d] becomes [g] because [t, d, n] are perceived by Amahl as velar before dark/velarized lateral (no coronal stops → velar before dark lateral rule)
Amahl’s production of puzzle with a [d] comes from a production problem, meaning…
[z] becomes [d] because fricatives get substituted by stops, which is a common process
Underapplication errors
Production rules apply across the board, while perception encoding problems only happen in contexts of perceptual confusion
Underapplication in Amahl’s production of puddle with a [g]
[t, d, n] are perceived as velar because of the raising of the tongue for the dark l, which masks the very short [t, d, n]
- There is no ‘across-the-board’ effect: [s/z] in ‘puzzle’- type words is realized as [t/d], but the derived [t/d] does not become [k/ɡ] because [s/z] are very salient
- A production rule would apply across the board (e.g. puzzle would be also pronounced with a [g])
Overapplication errors
When the [t, d, n] being perceived as velar is overcome, the correction [t, d, n] become no longer stored as velar before [ɫ]) will be applied incorrectly to some words
Steps of overapplication
- Puddle incorrectly stored as/pʌɡəɫ/
- Perceptual problem overcome : [g] now correctly perceived [d]
Problem : words with a velar like pickle can get incorrectly analysed as having a medial coronal(e.g. pitəɫ) until the word is heard again - Puddle now correctly perceived and produced as puddle
True or false : misperception is observed even in tasks with an optimal position for perceiving contrast (word-initial in stressed monosyllabic words)
True
when children show target-like discrimination in optimal positions, misperception can still happen in ..
Less optimal positions
In an experiment, French exposed 3 days old listened to bisyllabic strings,
Example: /mati/
mathématicien vs. panorama typique
Familiarization phase: Infants heard either within-word or cross-word instances of [mati].
Test phase: Switched from within-word [mati] to cross-word [mati], or vice versa depending on familiarization phase
What were the results of this study ?
Infants showed an increase in sucking rate when the within-word was switched to cross-word or vice-versa : novelty preference
- This means infants can discriminate between bisyllabic segments in a single word vs across words
What were the cues infants in the [mati] experiment used to infer word boundary ?
In a word boundary (ma#ti), the final vowel in the first syllable (/a/ in ma#) and the initial consonant in the second syllable (/t/ in #ti) are longer
In a study, 7 months old infants were presented with 2 versions of a story :
Version 1: with 1-second pauses in natural places (consistent phrase boundary cues)
o Version 2: with 1-second pauses in unnatural places (inconsistent phrase boundary cues)
What were the results of this study ?
Infants showed a familiarity preference : they listened longer to the version 1 with consistent phrase boundary cues
How stress can be used as a cue to word boundaries
If a language has mostly consistent stress, the word boundary aligns with or is close to the stressed syllable.
In English, 85% of ______ words have primary stress on the first syllable
Lexical (content words like nouns, verbs)
In an experiment, 7.5 months old were tested on their sensitivity to initiala stress
Familiarization: Infants familiarized on strong/weak words with primary stress or weak/strong words with final stress, across experiments.
* Testing: Infants tested on passages that either contained or did not contain the familiarized target words
What were the results of this experiment ?
7.5-month-olds detected the words only when they were strong-weak (initial stress) : listened longer to the passages containing these words.
* In passages with weak-strong final stress words like guitár that is always followed by “is”, infants missegmented táris as a word in a string like “guitár is” even after being exposed to guitar
Function words
- Express grammatical relationships, closed class, good word boundary indicators
o articles and determiners
o pronouns
o conjunctions
Lexical/content words
- Express content, open class
o nouns
o verbs
o adjectives
Why are function words segmentation cues ?
- Closed class
- Frequent
- Small (usually monosyllabic)
- Type of function word gives type of following word
How do children segment Det + N strings in French ?
Children try to identify CV determiners followed by C-initial nouns
Segmentation errors in French
Sometimes the syllable boundary does not align with the Det and Noun boundary
E.g. les arbres
[le.zaʀbʀ] does not mean that [zaʀbʀ] (zarbre) is a word
