Psycholinguistics Notes
Psycholinguistics
4/11/07
Methodology
Observational methods - slips of the tongue, etc.
Experimental methods:
Lexical Decision and priming - word recognition
Word-by-word reading, eye tracking - sentence processing
Brain Based methods:
PET, Functional MRI - brain activity scanned during a particular task.
ERP (event-related potential, p 443) - Measures brain activity during language processing. N400 spike-appears 400 ms after onset of word that doesn’t make sense. Even in non-native speakers. Syntax - 600 ms after a grammatical error, similar peak. This one fails for non-native speakers.
Findings:
Speech perception - Coarticulation - we don’t produce speech sounds individually. Sounds are influenced by neighboring sounds. How can we identify a sound when it has so many different variants in different contexts?
Categorical Perception - differentiate bw classes of sounds
VOT - voice onset time - when do vocal cords start to vibrate?
There’s a break bw 20-40 ms where sounds become differentiated. 60-80 ms is perceived as same sound. 0-20 perceived as same sound. 20-40 perceived as different.
Context - remove a segment from word, replace it w/ random sound, ask a native speaker what sound is missing. They’re very bad at it! They’ll think the random sound is just in background and nothing is missing.
Sentence context is also a very powerful cue in influencing how people perceive sounds.
McGurk Effect - we also use vision for speech perception. This is innate, even babies do this. Motor theory of speech perception - speech sound maps to a physical gesture related to producing the sound.
E-face and A-face experiment with infants: Showed faces first, no sound, then showed both faces together while playing sounds, to see if babies would look more at which one for which sound.
Word Recognition
Avg vocab size of a native speaker college-educated: 80K root words (not considering inflected forms). How do we search mental lexicon so fast?
- Frequency effect - higher frequency words require less time to recognize.
- Lexical status effect - takes longer to recognize a non-word as such
- Repetition - if word is presented twice, recognition takes less time the second time
- Priming effects (phonological, orthographic, semantic) - two words in sequence, recognition of 2nd word is faster if first is related. Even works subconsciously, if prime appears to quickly to consciously process (min 33 ms). Masked priming - hides prime behind hash marks or other letters, to fool your brain into thinking the prime appears longer.
Models of word recognition:
Serial Search model: 3 access files (visual, auditory, speaking) - analyze features of letters, compare visual input w/ lexical entry, activate “master file” in master lexicon. Words in lexicon contained in mental “bins,” w/ higher frequency words at top.
Interactive Activation Model: 3 levels of analysis: feature, letter, word - each represents a node (neuron); nodes are connected in various ways which trigger either activation or inhibition. Parallel process, activates or inhibits all words w/ or w/o a feature at the same time. Threshold: for a word to be recognized (node is fired), there is a threshold level of activation.
Sentence Processing
Garden Path sentences - lead you astray bc you want to make decisions about a sentence asap before reading the whole thing. “The horse raced past the barn fell” (verb is passive/transitive-horse that was raced). If verb can take an NP, we automatically associate the first NP we come across. Garden path model - incorporate elements into existing structure until you can’t. Relies on bottom-up processing. “John remembered (that) the doctor didn’t give him the prescription.” Sentence takes longer to interpret without “that.” Reduced relative clause. (Measured w/ self-paced reading task-each word is presented individually, subject presses space bar to move to next word.)
Context Effect - we use background knowledge to anticipate meaning of sentences. Don’t always do complete analysis of sentence. Constraint based model - uses other knowledge (background) in analyzing syntactic structure. Top-down processing.
Speech Production
Slips of the tongue
Speech errors most likely to occur w/ phonemes, morphemes, and words.
- Anticipation - anticipating next word, next phoneme. (I lent to the library)
- Perseveration - continue to use same phoneme (phonological fool)
- Exchange of phonemes or words (spoonerisms)
- Substitution of phonemes or words
- Blend (that’s torrible!)
- Movement of words in sentence (I have to do learn that)
- Deletion/insertion of syllables, phonemes (unamity of opinionion)
Stroop Effect: 1935
Words in diff colors - name the color of the word, not the word. Guess what, that’s hard!
· If color matches name of word - very fast.
· Color doesn’t match name of word - very slow.
· Color doesn’t match name of word, & word is not a color word = in the middle.
What does this mean? Identifying lexical aspect is faster, happens first, before non-lexical aspect. Recognition of a word is automatic, you cannot stop it, even if you tell people to ignore it. Once you recognize the word, the pronunciation becomes available, so recognition of word competes w/ output.