Chapter 12 – Psycholinguistics : the study of language processing

Chapter 12 – Psycholinguistics : the study of language processing 

psycholinguistics – the study of language-processing mechanisms; mental events

12.1 Methods of psycholinguistic research o o

language users are unaware of the details of language processing  subconscious  ability is maximized to operate with speed and efficiency saccades: a series of jerks in the movement of the eye while reading text  ≠ move smoothly from left to right

12.1.1 Slips of the tongue 

spoonerisms  slips of the tongue produced during speech - [field technique] o Reverend William A. Spooner  E.g. What he intended: Noble sons of toil What he said: Noble tons of soil o tendency to exchange the initial consonants of words in the utterance o sentences are planned out before the person begins to say it o ‘mixing and matching’ morphemes within words  E.g. Intended: easily enough Produced: easy enoughly o the morpheme, rather than the word, is the fundamental building block of English sentence production o morphological components of words can function independently during sentence planning

12.1.2 Experimental methods: words in the mind 

‘tip of the tongue’ phenomena  temporarily unable to access a word  reveals flexibility to access the mental lexicon



Lexical decision o participant is seated at a computer, a word appears on the screen and the participant must quickly judge whether the word is a real English word or not (‘yes’ or ‘no’) o two dependent variables  things that are being measured: the time that it takes for a participant to respond (response latency) and whether the participant’s judgment is correct (reponse accuracy) o frequency effect  participants take ½ a second (500 milliseconds) to press ‘yes’ for common words but take longer for less common words  suggests our mental dictionaries are organized so that words we use more often are more easily and quickly available to us o pronounceable non-words take longer to reject

 

aspects of phonology are automatically activated during word reading

The priming paradigm - target  word to be judged - prime  preceding stimulus o the extent to which the prime influences the participant’s lexical decision performance on the target stimulus is measured  response time is faster when a target is preceded by a semantically related prime (e.g. cat-dog)  networks  priming effect: when a word is seen, its representation is activated in the mind and spreads to other words in the lexical network that are semantically related; ‘warmed up’  orthographically related words – couch, touch  phonologically related words – light, bite  word roots and complex forms – legal, illegality

12.1.3 Experimental methods: sentence processing 

parsing  unconscious automatic analysis of the meaning of a sentence’s words and syntactic structure



Timed-reading experiments o Bar-pressing paradigm: participants are seated at a computer and read a sentence, revealing only one word at a time  dependent variable = the time it takes to process the particular word  participants do not show equal bar-pressing times across a sentence, but a pattern that reflects the syntactic structure of the sentence  bar-pressing times for processing content words (nouns and verbs) are longer  function words (determiners, conjunctions, and prepositions) are shorter  length of relative clauses and at the end of full sentences are longest Eye movements o when the eyes are at rest, they take a ‘snapshot’ of two or three words (lasting 200 to 250 milliseconds) o meanwhile, the language-processing system calculates where to jump to next o during a jump to the next fixation location (8 letters to the right), the reader is “essentially blind” o participant is seated at a computer displaying text, eye movements are tracked and recorded by a device illuminating the participants eyes with low-intensity infrared light  fixation times are typically longer for less frequent words  points of fixation are typically centred on content words (nouns and verbs) rather than function words (determiners and conjunctions)



o

regressive saccades: backward jumps in a sentence  usually associated with misparsing or miscomprehension (10-15%)  syntactically complex sentences and semantically anomalous sentences [The pizza was too hot to drink] create more regressive saccades

12.1.4 Brain activity: event-related potentials 

event-related potentials (ERPs)  produced by the brain, technique used in psycholinguistic research to measure electrical activity o similar to what participants do in normal language-processing situations o electrodes are placed on participant’s scalp, and recordings are made of voltage fluctuations resulting from the brain’s electrical activity as they read o difference between ERP and EEG (electro-encephalogram) is that in EEG, all electrical activity is recorded (from numerous background tasks) ERP is more advantageous because it calculates what part of the brain’s activity is related to a stimulus event  process of averaging  ERP pattern presented as a line graph; time = x axis, voltage = y axis, negative values on top, positive values on bottom  elicited by sentence-final words that are congruent, incongruent, and very incongruent  in the processing of implausible sentences, the brain displays a characteristic ERP sign of surprise  N400 – signal of semantic anomaly  P600 wave and ELAN (early left anterior negativity)

12.1.5 Language corpora and databases in psycholinguistic research  Researchers are able to analyze factors such as:    

age of acquisition # of syntactic context in which it can occur morphological family size semantic properties of the words that it has as neighbours in speech and writing o English Lexicon Project

12.2 Language processing and linguistics Phonetics and phonology 

language processing shows evidence that features, phonemes, and syllable structure all capture some aspects of the way in which we process language, but that speech production and perception is a complex activity that involves much more than these phonetic and phonological representations

 

bottom-up processing  inductive analysis to isolate phonemes and word boundaries, and to relate back to the mental lexicon top-down processing  employing a set of expectations to guide phonetic processing and word recognition o Features: the most basic level of representation  associated with bottom-up processing  characteristics of individual phonemes (e.g. [+voice], [-continuant],etc.)  slips of the tongue  big and fat, *pig and vat  feature change [voice]  /b-p/ and /f-v/ o Phonemes:  Cohort model – William Marslen-Wilson (1980’s): in word comprehension, words are analyzed by hearers from beginning to end o E.g. when we hear the word glass, we consider all words beginning with the sound [g], when the next sound is recognized, the # of possible words (the cohort) is reduced to words that begin with [gl] o this process continues until the cohort of possible words is reduced to one –the word that is being recognized  phoneme = the fundamental unit of auditory word recognition o Syllables:  role in speech perception  in normal auditory analysis, participants first break down stimuli into syllables, then into individual segments  in word-blending tasks, (bug + cat = bat), the first consonant of the 1st syllable is combined with the vowel and final consonant of the 2nd syllable  easier division with words using syllable structure suggests that words are represented in speakers’ minds syllabically

Morphological processing 

Morpheme activation o for most multimorphemic words, individual morphemes are automatically activated during word recognition o suntan primes semantic associates of its constituent ‘sun’ (e.g. moon) (1) multimorphemic words such as blackboard are represented in the mind as [black + board] respectively when we access such words, both their whole word forms and constituent morphemes are automatically activated  post-lexical decomposition [black] [board]  [blackboard] (2) activation of constituent morphemes results from a computational mechanism (morphological parsing) that scans a word and isolates individual morphemes 



pre-lexical decomposition [blackboard]  [black] [board] o bark: 1. ‘the outside of a tree’, 2. ‘the sound a dog makes’ barking  both semantic associates were affected during pre-lexical decomposition bark and –ing were accessed separately, allowing both meanings of bark to be activated before one was suppressed  words that look multimorphemic but are not can be primed in the same way (e.g. corner  corn and the suffix –er (much in the same way as hunter primes hunt)  multimorphemic words that do not look like root + suffix combinations do not show such effects (scandal ≠ prime scan) Selectional restrictions o processing times for illegal nonsense words (‘rebirmity’) were significantly longer than for morphologically legal words (‘rebirmable’, ‘rebirmize’) o due to the prefix and suffix rule that re- must attach to a verb and -ity must attach to an adjective, yet the root ‘birm’ cannot be a verb and an adjective at the same time  suggests that knowledge of the selectional restrictions of affixes does indeed form part of the word-processing system and that violation of these restrictions create difficulty for automatic lexical processing

Syntax  

sentence formation arrangement of words into a syntactic structure is unique to each sentence o The syntax module  speakers begin with deep structure representations and employ a series of transformations to derive the surface structure characteristics of a sentence  the # of transformations in a sentence did not predict processing time  syntactic parser  a special module for sentence processing and grammatical knowledge  module  a mechanism of processing that is relatively autonomous from other processing mechanisms o Garden path sentences  sentences that are difficult to understand even though they are not complex syntactically (‘The horse raced past the barn fell’) o perfectly grammatical o difficult to understand (1) Minimal attachment: we prefer not to postulate new syntactic nodes (2) Late closure: we prefer to attach new words to the clause currently being processed as we proceed through a sentence from beginning to end

o

 Parsing system is a module that operates automatically and independently Sentence ambiguity o They all rose.  rose  a flower /  to stand  after seeing the sentence, participants were presented either the word flower or stand, both were activated even though the sentence clearly presented a bias in favour of one reading over the other o suggests that sentence processing proceeds in two stages: 1. all possible representations and structures are computed 2. one of these structures is selected and all others abandoned

12.3 Putting it all together: psycholinguistic modelling 







The use of metaphors in psycholinguistic modelling o dual functions  summarize specific research findings and generate specific hypothesis  important function of embodying general perspectives on how language processing works  metaphors have the effect of shaping how we conceive of language in the mind and what kinds of questions are asked by psycholinguistic researchers  they provide the means by which major families of models can be contrasted in order to test which ones most accurately describe language processing Serial versus parallel processing models o serial processing model – claims that language processing proceeds in a step-by-step manner  modular, bottom-up (i.e. phonetic perception) o parallel processing model – claims that phonological, lexical, and syntactic processes are carried out simultaneously  complex processes (i.e. sentence comprehension) Single-route versus multiple-route models  single-route models – claims that a particular type of language processing is accomplished in one manner only  multiple-route models – claims that a language-processing task is accomplished through (usually two) competing mechanisms Symbolic versus connectionist models o symbolic models – claims that models of linguistic knowledge must make reference to rules and representations consisting of symbols, such as phonemes, words, syntactic category, labels, etc. o connectionist models – claims that the mind can be best modelled by reference to large associations of very simple units (called nodes) that more closely approximate the kinds of processing units (i.e. neurons) that we know the brain to be composed of