Auditory Word Recognition Flashcards
Introduction - Auditory Word Recognition
- Spoken words difficult. Brief. Small pauses.
Although difficult, people good at recognising speech, automatic process which is effortless and quick.
Marslen-Wilson (1984) - Ppl understand sentences spoken at more than 20 phonemes per second.
Explain how variability is a problem when understanding speech sounds.
Idea that the same phonemes can sound differently in different contexts.
Assimilation - Phonemes can be influenced by some acoustic properties of their neighbours due to co-articulation. Details of the realisation of a phoneme can vary depending on context surrounding.
Co-articulation = As make one sound, vocal apparatus move, preparing to make another sound. Produces speech more quickly.
- Speaker variability - No two speakers produce the same phonemes in the same way e.g. accents, physical properties of the mouth.
Explain how segmentation is a problem when understanding speech sounds
Words uttered without a break between e.g. ‘I scream’ and ‘ice cream’ sound similar.
Norris et al. (1997) - The possible-word constraint is a strategy of segmenting speech in such a way that there is no syllable left unattached to a word
Another strategy to segment speech depends on an indvs exposure to language.
Cutler and Butterfield (1992) - The metrical segmentation strategy of speech involves an insertion of word boundaries before syllables and thus eliminate weak ones. Stress based segmentation can vary between languages. E.g. when hear ‘conduct ascents uphill’ they heard ‘the doctor sends the bill’
Liberman et al. (1957)
Participants presented with continuum of artificially synthesised syllables and the task was to report what they heard.
/ba/-/pa/ with different VOT times
- Syllables with 0msec VOT as /ba/
- 60msec as /pa/
- 20-40msec as /ba/ or /pa/
Syllable perception was affected by the Voice Onset Time
Describe what the McGurk effect is
A perceptual illusion which shows that senses work together. Individuals perceive the mouth moving as pronouncing /ga/. The sound is /ba/ but people report hearing /da/.
Shows that phoneme perception can be influenced by input from another modal
Describe the dual-code model of phoneme perception
- Foss & Blank (1980)
- Suggests that there are two ways for phoneme perception and that speech perception uses a prelexical code and a post-lexical code.
- Pre-lexical acoustic information
- Code computed directly from perceptual analysis of the input.
- Used for easy tasks or processing of non-words - Post-lexical
- Code derived from info derived from higher level units such as words
- Difficult tasks
Explain evidence which supports the dual-code model of phoneme perception
Foss & Blank (1980)
- Phoneme monitoring task
- Phoneme monitoring times to target phonemes in words and nonwords mostly the same.
- Participants responding to phonetic code as nonwords cannot have phonological codes.
- Respond to prelexical when task is easy and post when task difficult.
Foss & Gernsbacher (1983)
- No support for model
- Used phoneme monitoring task using multiple targets
- Increasing processing-load does not produce effects based on the post-lexical code, suggests that the post-lexical code may be used but only in very exceptional circumstances
Marslen-Wilson and Warren (1994)
- Lexical decision task to question if phonemes have to be identified before spoken word recognised
- Start with word smog and then take off final consonant and replace with b to get smob.
- Smog more difficult to reject than nonword because co-articulation info from the vowel is consistent with a word.
- Would have been equally difficult if phonetic rep of the vowel had been translated into a phoneme before lexical access
- Lexical reps are accessed from featural info in the sound signal and co-articulation is used early to identify the following consonant
Warren (1970)
- Study looking at the timing of the use of context in auditory word recognition.
- Shown “the state governors met with their respective legi*latures covening in the capital city”
- is when word cut and replaced with cough
- Participants could not detect that a sound was missing and they restored the /s/ to the word
Warren and Warren (1970)
- Study looking at the timing of the use of context in auditory word recognition.
- Shown 4 different sentences in which the only thing different was the last word so no phonological diffs that might cue
- The phoneme restored depended on the semantic context given by the final word
- Shows that people are good at restoring sounds using context
Samuel (1981, 1987…)
- Study looking at the timing of the use of context in auditory word recognition.
- Examined adding noise to the segment instead of replacing with noise
- Lexical context does lead to true phoneme restoration and that effect was prelexical
- Sentence context does not affect phoneme recognition and only affects postlexical processing
Describe the Cohort model by (Marslen-Wilson, 1984)
As we hear speech, we set up a cohort of possible items that the word could be and then items are eliminated until only one left in cohort.
3 stages
Access - Acoustic features are analysed and cohort created
Selection - One item chosen from set
Integration - Syntactic and semantic properties of the word are used. Higher level info used e.g. does meaning fit with context.
Distinguishes between diff points in time
- Uniqueness Point - Point at which a word becomes uniquely identifiable from initial sound sequence
- Recognition Point - Point at which word is empirically identified
Zwiterslood (1989)
- Testing the cohort model using cross-modal priming with lexical decision
- Found frequency effects in cross-modal priming, but the context effects did not show until later stages of processing
Describe the revised cohort model (Marslen-Wilson, 1987, 1989)
- Context only affects the integration stage and cannot be used to eliminate members of the cohort before the uniqueness point
- Bottom up priority - Context used at a later stage as cannot be used to form initial cohort ]
- Frequency effects - Word frequency affects the activation level of candidates at early stages
- No ‘all or none’ elimination - Incompatible bottom-up evidence partially eliminates a candidate
- Reactivation of decayed candidates - Gradually decays but can be activated later when further evidence available
Evaluate the cohort model
Frequency effects - Revised assumes that high frequency words get activation quicker than low frequency cohorts from same evidence which can explain early frequency effects.
Neighbourhood size - Model predicts the size does not affect recognition time of the word as the activation takes place in parallel across all candidates - can’t explain the effects
Neighbourhood frequency - The model assumes different activation rates for different frequencies, can.
Context effects - Context influences post-lexical processing only - can explain late context effects
