Understanding words

approx. 75,000 words in memory
1. stored in mental lexicon (in LTM)
  1. contains lexical entries, one per word, meaning spelling pronunciation etc
  2. lexical access is retrieving a word from mental lexicon
    1. seeing/hearing a word activates lexical entry
      1. if activation reaches threshold access takes place i.e realise you're talking about toast
    2. Gap between recognising word (accessing representation in ML) and accessing meaning called 'magic moment' Balota (1990)
Words must be organised in systematic way, find out how by looking at things that make lexical access easy or hard
1. Lexical decision task: participants decide whether a string of letters is a word or nonword. Reaction time and error rates measured. Varient of this-lexical naming
  1. Problem: speed accuracy trade off (faster = mistakes) careful instructions help this. Also within measures better individual differences in quickness of reading
2. Word length: Gough (1972) in word recognition letters are taken out of memory one by one at rate of 15ms, not surprising longer words slower to access than short
  1. Also with numbers Klapp (1974)
  2. Problem: word length effect that is independent of frequency has proved elusive, diff ways of measuring word length syllables, letters how long it takes to say. Whaley (1978) found some word length affects lexical decision BUT Henderson (1982) did not find this. Chumbley & Balota found effect when words and nonwords were match for length and regularity of pronunciation.
    1. Naming time increases as a function of syllables in the word Eriksen, Pollack & Montague (1970). Some contribution from preparing to articulate these syllables in addition to any perceptual effect
3. Word frequency: common words easier to recognise and respond to quickly
  1. first shown in briefly presented word recognition Howes & Solomon (1951)
    1. Whaley (1978) frequency most key factor in determining speed of responding in lexical decision task. Even when words are same length, frequency effect Rubenstein et al (1970)
      1. Forster & Chambers (1973) frequency effect in naming task
        Therefore essential to control for this effect in psycholinguistic task. List of frequency available Kucera & Francis (1967) BUT only approximation some low frequency words high to professionals Geinbascher (1984)
4. Context: general way to talk about priming. Priming from sentence context is over and above that of the associative effects of individual words in the sentence. Words are recognised better in sentence contexts
  1. facilitates recognition even though there is no semantic relation between words ie day & teeth
  2. Schuberth & Eimas (1977) context effects in visual word recognition: presented incomplete sentences followed by word or nonword to which participant made lexical decision. Response times were faster if the target word was congruent with the preceding context . Sentence context can have either an early perceptual effect or a late post perceptual effect.
  3. Lieberman (1963) heard words either in isolation or sentences, isolation took 2X longer to recognise
  4. Bruce (1958) heard words against background noise, better recognition in context
5. Priming:when language processing is facilitated by prior context
  1. once a word is identified its easier next time, lowers activation threshold.
    1. Repetition priming: facilitates both the accuracy of perceptual identification (Jacoby & Dallas, 1981) and lexical decision response times (Scarbourgh et al, 1977). Has long lasting effect hours or longer. It interacts with frequency, effects are stronger for low frequency words known as frequency attenuation (Forster & Davis, 1984), they also found it has brief lexical access effect and LT episodic effect
      1. Debate as to whether RP occurs due to activation of items stored representation (Tulving & Schacter, 1990) or because of creation of record of entire processing in episodic memory (Jacoby, 1983)
      2. Important supporting episodic view: we generally obtain facilitation by RP only within a domain (i.e. visual or auditory modality), BUT semantic priming work across both domains Roedige & Blaxton (1987).
  2. Semantic priming: identification of word facilitated by prior exposure to a word related in meaning (Cattel 1947)
    1. lexical decision making task Meyer & Schvendeveldt (1971) identification of nurse easier/faster if preceded by doctor. first word 'prime' second word 'target'. In short time intervals priming can occur even if prime follows the target (Kigert & Glass, 1983)
    2. type of context effect, can see the effect might have some advantages for processing. words related in meaning sometimes co-occur in sentences. Hence processing may speed up if words related to the word you are reading are somehow made easily available, as they are more likely to come next than a random word
  3. Spreading activation model accounts for priming effects: think of words in LTM as a network, multidimensional connections (meaning, context, sound)
    1. accessing a word causes activation, activation spreads to other nodes
  4. Phonological priming: Evett & Taylor (1982) words related in sound, share pronunciation, faster. not only semantic network but also phonological network
6. Lexical ambiguity: when a word has more than one meaning
  1. 'bank' general bias in favor of financial. work our bias in order to test lexical access
    1. Frazier & Rayner (1990) distinguished between words with multiple unrelated meanings and words with multiple senses (twist) which are related.
  2. homographs: written the same but have different meanings (not ambiguous when spoken)
  3. How do you choose right meaning?
    1. Context helps select relevant meaning. HOW? Immediately or after consideration of all meanings? 3 models of context effects
      1. Autonomous access model (modular): all meanings accessed independent of context and context is later used to select correct meaning in integration phase. Sub processes are different domains and info can be integrated later on
      2. Re-ordered access model: Get all meanings but speed at which you get one is affected by context. availablity of appropriate meaning is increased (Patch & Duffy 1994) influence context can have is limited
      3. Direct access model (interactive): Only contextually appropriate meaning is accessed, v specialised machinery, info at one domain used immediately for another. Unclear how context can provide immediate constraint
      4. Swinney (1979) cross-modal lexical decision bugs: insect/listening device. context (spiders) supports insects BUT do people initially consider both meanings? hear ambiguous sentence and decide if word (ant/spy/sew) shown after they hear bugs, is a word. fast at first two, shows both accessed even though only ant was contextually relevant. Time to access tells us how easy retrieval is. When word is shown after insects (unambiguous sentence) ant is fast spy is slow. So both meanings were immediately available after bugs - context had no immediate effect
        Under normal circumstances (unambiguous) you get priming because insect is related to ant BUT bugs is ambiguous so both activated, context not yet been used to distinguish between 2 meanings
        swinney also tested 3 syllables further in ambiguous sentence (later than bugs) you activate all meanings after bugs but 3 syllables after you are only left with context appropriate meaning. Suggests lexical access makes all meanings available regardless of context but context then kicks in rather fast (integration). In favor of Modular model: 2 systems meaning THEN context.
        is this always true?
        Processing can vary depending on the particular ambiguous word and the particular prior context. All of swinneys words were balanced (equally common) what about biased words?
        Onifer & Swinney (1981) one meaning of word was more frequent than other meaning, yet they still found that all meanings were activated regardless of biasing context. However dominant meaning may be accessed more strongly and perhaps sooner than less frequent ones (Simpson & Burgess, 1985)
        However: Rayner & Duffy (1986) used eye tracking (gaze duration indicates difficuclty of lexical access) to find that biased words people accessed only 1 meaning (more common) easier to get correct meaning and with balanced words both meanings are accessed due to the two competing, longer gaze/harder to get correct meaning.
        Duffy et al (1988) manipulated whether the disambiguating context came before or after the critical word. Also manipulated whether ambiguous word was balanced. Context after lexical access was difficult for balanced = lexical access happens before disambiguating context, 2 meanings compete. Context helps. Context before lexical access harder for biased, context promotes less common meaning and this competes with activation of more common meaning always retrieved. Context hinders.
        This is called Subordinate Bias effect: context before can re-order , meanings so both dominant and subordinate available at same time additional processing needed to select appropriate subordinate meaning (Rayner et al 1994): 2 accounts
        Integration account: reach dominant meaning first for 'port' but doesn't fit with context at integration stage so go back and retrieve other subordinate meaning, takes times hence difficulty
        Re-ordered access account: dominant meaning accessed at same time as contextually relevant (subordinate) meaning, need time to resolve competition
        Dopkins et al (1992) wanted to distinguish between 2 accounts, used eye-tracking. Positive condition used prior context info to speed up selection of uncommon meaning. Does not support integration model, would have predicted difficulty at the end of sentence due to more common selection of meaning early on then integration of disambiguating context activating uncommon meaning. As there was less difficulty this supports re-ordered access account increasing availablity of less common/apppropriate meaning due to inital phrase
        Context sensiitive model (Martin et al 1999) activation of word meaning affected by: meaning frequency, type of biasing context (dominant etc), contextual strength. The subordinate-bias effect (content supports less common meaning) should disappear in a strongly constraining context as frequency of meaning should have weaker effect
        Used self-paced reading to test this. Biased words with contexts supporting dominant or subordinate meanings.The strength of context was manipulated. V strong context no difficulty with less common meaning, with weak there is difficulty. Supports model and argues you can remove subordinate bias effect

Next up

Understanding words

Description

Resource summary

Similar

	Created by issy_hinds about 11 years ago