The Lexicon and Lexical Access
The Lexicon
- Dictionary versus Encyclopedia
- Lexical Entries
- HEARSAY’s Lexicon
- What is word meaning?
Dictionary versus Encyclopedia
- The lexicon contains everything we know about words.
- The lexicon is described as a "mental dictionary".
- It is not the same as semantic memory, which is more like a mental encyclopedia.
- Some Reasons for the Distinction (Clark & Clark, 1978)
- Not all concepts have names.
- Not everyone speaks a language.
- Which came first? Do we really need a lexicon?
Lexical Entries
- Each lexical entry represents a word.
- Each entry contains information about:
- sound pattern
- spelling
- syntactic category
- meaning
HEARSAY’s Lexicon
- HEARSAY has a lexicon of 31 chess words.
- Each entry includes information about:
- sound pattern
- syntactic category
- meaning
What is word meaning?
- The Information in the Lexicon’s Meaning Slot
- A Set of Semantic Primitives
- A Location in Semantic Space
- A Set of Procedures
- Its Relationships to Other Words
The Information in the Lexicon’s Meaning Slot
- Encyclopedia Entry
- Critical Components
- Nominal View -- Pointer to Semantic Memory
- These are fuzzy distinctions that don’t really address the issue.
A Set of Semantic Primitives
- Examples
- bachelor = [-female] [+adult] [+human] [-married]
- This accounts for a variety of phenomena.
- Analyticity: "That woman is female."
- Self-Contradiction: "That woman is male."
- Synonymy: "couch" versus "sofa"
- Antonymy: "woman" versus "man"
- Problems
- What are the primitives?
- Can primitives represent all concepts (e.g., "red," "squirrel")?
- Some meanings require continuous primitives (e.g., "big").
A Location in Semantic Space
- Multidimensional Scaling
- Relationship to Components
A Set of Procedures
- Example: The Meaning of "Man"
- Step 1: Is X a human?
- Yes: Go to Step 2
- No: Go to Step 5
- Step 2: Is X an adult?
- Yes: Go to Step 3
- No: Go to Step 5
- Step 3: Is X male?
- Yes: Go to Step 4
- No: Go to Step 5
- Step 4: The procedure succeeds, X is a man.
- Step 5: The procedure fails, X is not a man.
- Relationship to Procedural Definitions
- Relationship to Components
- Representation versus Process
Its Relationships to Other Words
- Semantic Network
- This is the only viable alternative to components.
Lexical Access
- Methods
- Properties of Lexical Access
- Demonstrations
- A Computational Analysis of Lexical Access
- Models
- Special Cases
- Lexical Access and the Brian
Methods
- Lexical Decision Task
- Naming Task
- Fixation Time
Properties of Lexical Access
- Automaticity
- Frequency Effects
- Context Effects
- Orthographic-Phonological Regularity
- "must"/"like" versus "have"/said"
- Frequency X Regularity
A Computational Analysis of Lexical Access
- What information is available?
- What is the goal of the computation?
- What strategy is used to achieve the goal with the available information?
- Search Models
- Threshold Models
- Explaining Frequency Effects
Models
- Forster (1981)
- Becker (1979)
- The I. A. M. (McClelland & Rumelhart, 1981)
- Seidenberg & McClelland (1989)
Forster (1981)
- A Search Model
- Assumes a GPS along with independent lexical, syntactic, and semantic modules.
- Language processing is strictly bottom-up.
- Frequency effects are due to order of search.
- Context effects are attributed to "post-lexical" decision making that occurs
in the GPS.
- How about orthographic-phonological regularity?
Becker (1979)
- Another search model.
- Assumes four stages:
- Generate Semantic Set
- Search Semantic Set
- Generate Sensory Set
- Search Sensory Set
- Frequency effects are due to search order.
- Context effects occur because the semantic set is generated first.
- How about orthographic-phonological regularity?
The I. A. M. (McClelland & Rumelhart, 1981)
- An example of a threshold model.
- Frequency effects occur because high frequency words have higher baseline levels of
activation.
- Context effects might be explained by implementing additional levels of representation
and by allowing activated words to "settle" above their normal baseline levels
of activation.
- Regularity effects might be explained by implementing the "full" model.
Seidenberg & McClelland (1989)
- The Conceptual Model
- The Implementation
- Representational Assumptions
- Processing Assumptions
- Evaluation
The Conceptual Model
The Implementation
Representational Assumptions
- 400 Orthographic Units
- Distributed Representations of Visual Features that Make Up Letter Triplets
- MAKE = _MA + MAK + AKE + KE_
- 460 Phonological Units
- Distributed Representation of Phonological Features that Make Up Phoneme Triplets
- /mAk/ = _mA + mAk + Ak_
Processing Assumptions
- The model is trained using the back-propagation learning rule.
- Processing is Parallel and Cascaded
- Error scores are used to estimate settling times.
- Naming Times
- Lexical Decision Times
Evaluation
- Predicts faster reading times for high frequency words.
- Top-down effects of context and meaning might explain context effects if they were
implemented.
- Neighborhoods explain the effects of orthographic-phonological regularity.
- Predicts developmental trends.
- Predicts effects of developmental dyslexia.
- Predicts naming times for non-words!
- Could a search model do that?
- Where’s the lexicon?
Special Cases
- Ambiguous Words
- Morphologically Complex Words
Ambiguous Words
- Background
- Possible Models
- Swinney’s (1979) Cross-Modality Priming Study
- Predictions and Results
Background
- Ambiguous words have more than one meaning.
- A "bug" can be an insect or a listening device.
- A "bank" can be a financial institution or the side of a river.
- One meaning (the "dominant" meaning) is usually much more common than the
other.
- How do we access the appropriate meaning of ambiguous words?
Possible Models
- Context Dependent
- We only access the meaning that is appropriate in the context.
- Ordered Access
- We access the dominant meaning first. If it is inappropriate we then access the
non-dominant meaning.
- Exhaustive Access
- We access all of the meanings in parallel.
Swinney’s (1979) Cross-Modality Priming Study
- A story is presented over earphones.
- "Rumor had it that for years the government building had been plagued with
problems. The man was not surprised when he found several spiders and other bugs ^ in the
^ corner of his room."
- A probe word (ant/spy/sew) is presented visually immediately or after two syllables.
- Participants make a speeded lexical decision judgement.
Predictions and Results
- Priming
- Semantic associates of activated meanings will be recognized more quickly.
- What are the predictions?
- For the context dependent model?
- For the ordered access model?
- For the exhaustive access model
- What are the results?
- Which probe words are primed immediately?
- Which probe words are primed after two syllables?
Morphologically Complex Words
- Morphemes are the smallest units that carry meaning or syntactic information.
- Morphologically complex words are made up of multiple morphemes.
- "unhappy" = "un" + "happy"
- "infrequently" = "in" + "frequent" + "ly"
- "dogs" = "dog" + "s"
- How are morphologically complex words represented?
- Like any other word.
- With separate lexical entries for each morpheme.
- Can you design an experiment to determine which is right?
Lexical Access and the Brain
- Neural Style Computation
- The 100 Step Limit
- Symmetry of Neural Connections
- Patients with Wernicke’s aphasia and deep dyslexia often make meaning based word
substitutions ("horse" --> "cow").
- Can any of the models we have looked at explain this?
- If we were to implement Seidenberg & McClelland (1989) entire conceptual model do
you think it could explain this?
The End!
|