Reading and Speech Perception
Reading
Speech perception
Defining language
A shared symbolic system for communication
Linguistics
The discipline that takes language as its topic
Psycholinguistics
The study of language as it is used and learned by people
Reading vs Speech Perception
Understand sentence in same way whether we read text or listen to someone talking
However, crucial differences
click to edit
Reading
Words seen as a whole
Low ambiguity
Rarely distracted by other stimuli
Low cognitive demands
Punctuation main cue
click to edit
Speech Perception
Words spread-out over time
High ambiguity
Adverse conditions in everyday life
High cognitive demands
Prosodic cues
Involve different brain areas (evident from brain damage)
Reading processes
Reading involves several different kinds of processing
click to edit
Orthography (word spelling)
Phonology (word sound) – phoneme being smallest unit of sound, and morpheme being a combination of sounds that make a word
Semantics (word meaning)
Syntax and grammar
Higher-level discourse integration – whole texts
Research methods
Several methods available for studying reading
Naming task
Say printed word out loud as rapidly as possible
Link orthography (spelling) and phonology (sound)
Lexical decision task
Decide rapidly whether string of letters from a word
Link orthography (spelling) and semantics (meaning)
Prime word task
Say the printed word out loud as rapidly as possible
Does a word presented before a target word affect processing of the target?
Understand and address reading difficulties
Majority of studies consider English language
Angloscentricites: relationship between spelling (orthography) and sound (phonology) inconsistent
English children learn to read more slowly than children learning a more consistent language (Caravolas et al., 2013)
Phonological processing
Do we access sounds when reading words?
Weak phonological model
Phonological processing is inessential for word identification
Strong phonological model
Phonological processing central for word identification
Evidence consistent with strong model
Homophones
Words with one pronunciation, but two spellings, e.g. is it a flower? rows vs robs
More errors made when word is homophone of real word (e.g. ROSE)
Errors suggests engaged in phonological processing
Phonological neighbours
Words that differ in one phoneme
e.g., GAIT - BAIT - GET
When reading a sentence, look at words with many neighbours for shorter amount of time
Advantage suggests engaged in phonological processing
Phonological priming
Words processed faster when prime is phonologically identical
e.g., klip vs cliip & parb vs clip
Advantage suggests engaged in phonological processing
However, phonological processing may not be essential for effective reading
Brain damaged patients can have impaired phonological processing, but still understand the meaning of words
Word recognition
Interactive activation model of visual word processing
Recognition units at three levels
click to edit
Feature level
Letter level
Word level
Involves parallel processing
Bottom-up and top-down processes interact
Interactive activation model
Model can account for the following
Word superiority effect
Target letter readily detected in a letter string when the string forms a word rather than a non-word
Orthographic neighbours
Words formed by changing one of a target word’s letters and can influence recognition
Emphasis on top-down processes from word level to letter level
Word superiority effect
Letter string presented briefly followed by a mask
Letters from the string are presented in a different order, but one is missing
Asked what letter was in a given position (e.g. for CAP, P A is shown, asked what letter was in the third position – the answer is P)
Better when letter string forms a real word (because a non-word can be used)
Orthographic neighbours
Words formed by changing one letter in target word
target: STEM
orthographic neighbours: SEEM STEP STEW
When word presented, neighbours activated
Neighbours facilitate target word recognition if they are less frequent (weakly activated) in language
Neighbours inhibit target word recognition if they are more frequent (strongly activated in language)
Interactive activation model
Limitations
click to edit
Too much importance attached to letter order
Should have problems reading the following
Able to read text when first and last letters in correct place but others transposed
Msot of you wlil pbroblay hvae no dfifcitluy redanig tihs eevn touhgh the lteerts are all mxeid-up!
Semantic priming
Target word recognised faster is preceded by semantically related word
prime: NURSE target: DOCTOR
Two possible explanations
Automatically activates related words due to learning
Expectation semantically related word will follow
Reading aloud
Two major theoretical approaches
Dual-route model (Coltheart et al., 2001)
Two routes between printed word and speech
Activation at one level cascades onto the next
Connectionist triangle model (Plaut et al., 1996)
Based on highly interactive system between orthography (spelling), phonology (sound) and semantics (meaning)
Semantics plays greater role in reading aloud
A complex process that involves various stages
click to edit
- Select signal of interest from irrelevant inputs
- Extract (or decode) the elements of interest (e.g. phonemes) from the speech
- Word identification
- Comprehension and interpretation
Construct meaning and integrate to construct speaker’s intended message
Listening to speech
Must deal with considerable variability in signal
Adverse conditions decrease intelligibility
click to edit
Phonemes pronounced in different ways (e.g. accents, dialect, rate of speaking)
Energetic masking: target degraded in some way (several people talking at the same time)
Information masking: effect of cognitive load (e.g. completing multiple tasks)
Dealing with variability
Stress
Initial syllable of content words (nouns, verbs) typically stressed in English
Coarticulation
Pronunciation of phoneme depends on preceding and following phonemes
Sentence context
Information that is not provided in the auditory signal
Sentence context
Information that is not provided in the auditory signal includes
Information provided by previous input (e.g. earlier parts of a sentence)
Top-down information provide by knowledge and experience of language/words
Has a rapid influence on speech perception
Phonemic restoration effect
Listener unaware that a phoneme has been removed and replaced by a non-speech sound
The state governors met with their respective legi*latures convening in the capital city
*cough (but will be perceived as a “s” sound)
Evidence that we use top-down expectations based on sentence context
Listening to speech
Speech perception is a complex process that involves various stages
Must deal with considerable variability in signal
Use various cues to help us deal with variability
click to edit
Stress
Coarticulation
Sentence context
Theories of speech perception
Theoretical accounts of the processes involved in spoken word recognition
Three major theories
click to edit
Motor theory (Liberman et al., 1967)
TRACE model (McClelland & Elman, 1986)
Cohort model (Marslen-Wilson & Tyler, 1980)