Word recognition Flashcards
Womb learning who?
Moon et al. 2013
Womb learning what?
30 hour old babies were showed vowels in their mothers language and vowels in other languages
Womb learning what language was mothers tongue
Swedish or english
Womb learning findings
babies sucked their dummies harder when listening to a foreign language, shows they learn some speech in the womb
characteristics of what makes word recognition difficult
segmentation problem
invariance problem
segmentation problem
speech is a continuous stream of sound with no white space between words.
segmentation problem (2) solutions
possible word constraint (norris et al 1997)
Stress (cutler and butterfield 1992)
possible word constraint (norris et al 1997)
Fapple vs Wuffapple, any segmentation that results in impossible words is likely rejected (eg fapple)
Stress (cutler and butterfield 1992)
first syllable is typically stressed. Conduct ascents uphill hears as doctor sends the bill
The invarience problem
speech sounds are shaped by external factors
Co-atriculation
letters can be pronounced different such as e in mercedes or b in bill ball bull and bell.
“lean, bacon, lean bacon”
what causes coarticulation
dialect, speech rate and accent
context effects
knowledge about the contect in which a word occurs can help word recognition
semantics, syntax, etc.
Phenome restoration effect who?
warren and warren 1970
Phenome restoration effect what
audio on slide 10; was there a phenome missing “the state governers met with their respective legislatures convening in the capital city.”
Phenome restoration effect example (2)
It was found that the *eel was on the orange
It was found that the *eel was on the axle
It was found that the *eel was on the shoe
It was found that the *eel was on the table
peel, wheel, heel, meal
theories of spoken word recognition
TRACE AND COHORT - most prominenet models
TRACE who
McClelland and Elman 1986
TRACE what?
3 layers - word units, phoneme units, feature units,
Q: How does information flow in the TRACE model?
A: Information flows both bottom-up (from features to words) and top-down (from words to features).
Q: What happens during word recognition in the TRACE model?
A: Words compete for recognition based on their similarity to the input, and activation levels change dynamically as speech unfolds.
Q: How does context influence word recognition in the TRACE model?
A: Higher-level context can activate or suppress certain word candidates, influencing recognition.
Q: How does the TRACE model handle continuous speech?
A: The model recognizes words in continuous speech without explicit word boundaries, using dynamic competition and activation processes.
A criticism of TRACE
TRACE overestimates the influence of context
and predicts top-down effects where they do not
exist
A criticism of trace cont.
Frauenfelder et al. (1990): TRACE predicts it’s hard to
detect /t/ in (French) nonword vocabutaire because
of the similar real word vocabulaire. Data did not
support prediction.
Cohort model who?
Marlsen Wilson 1984
Q: What is the “initial cohort” in the Cohort Model?
A: The initial cohort is the group of words activated based on the initial sounds of the spoken word.
Q: How does the Cohort Model narrow down possible words?
A: As more speech input is processed, words that no longer match the input are eliminated from the cohort.
Q: What is the “uniqueness point” in the Cohort Model?
A: The uniqueness point is the moment when only one word matches the incoming speech input.
Q: How does context affect word recognition in the Cohort Model?
A: Context can influence cohort activation or elimination by providing semantic or syntactic cues.
Q: How does the Cohort Model differ from the TRACE model in processing?
A: The Cohort Model uses a step-by-step narrowing of candidates (serial processing), while TRACE uses interactive, parallel processing.
Q: How does the Cohort Model handle continuous speech?
A: It continuously updates the cohort as speech input unfolds, parsing overlapping signals in real time.
Q: What factors influence word selection in the original Cohort Model?
A: Word selection is influenced by the auditory input and semantic/syntactic context
Q: How can context affect word recognition in the original model?
A: Context can allow a word to be recognized before its uniqueness point by supporting a single candidate in the cohort.
Q: What issue does the original model face with mispronunciations?
A: Words with mispronounced initial phonemes (e.g., “focabulaire” for “vocabulaire”) should not be recognized, but evidence shows they can still be.
Q: How does the revised Cohort Model handle semantic context?
A: Semantic context no longer influences the early stages, making the model more bottom-up.
Q: How are words managed in the revised model if they mismatch the input?
A: Words are not eliminated but have reduced activation, allowing revival if new evidence supports them.
flashcard 6: Evidence for the Revised Model
Q: What evidence supports the revised model’s handling of mispronunciations?
A: Hearing “focabulaire” can still activate “vocabulaire” (Frauenfelder et al., 2001), showing mispronounced words are not fully excluded.
Q: What did Zwitserlood (1989) find about semantic context in the revised model?
A: Semantic context does not influence which words remain in or leave the cohort of candidates.
