Lecture 8: Word Meaning, Inference, and Reference Study Notes

PSYC 145: Lecture 8 - Word Meaning, Inference, and Reference

Word Meaning and the Mental Lexicon

  • Conceptual Recap: Semantic Priming

    • Word meaning is fundamentally represented within the mental lexicon.

    • Words are structured within this lexicon through some form of interconnectivity.

  • The Semantic Verification Task

    • This task involves participants judging whether a specific statement is "true" or "false."

    • Research conducted by Collins & Quillian (1969) utilized this task to measure response times associated with different types of semantic relationships.

    • Examples Regarding Categories:

      • "A robin is a robin." (Fastest response time).

      • "A robin is a bird."

      • "A robin is an animal." (Slowest response time).

    • Examples Regarding Attributes:

      • "A robin has a red breast." (Fastest response time).

      • "A robin has wings."

      • "A robin has lungs." (Slowest response time).

Hierarchical Semantic Network Model

  • Structure and Organization:

    • Concepts are stored in a strict hierarchy. For instance, "animal" is positioned at a higher level than "bird."

    • Property Inheritance: Properties are stored at the highest relevant node to save cognitive space. Lower-level nodes inherit these properties. For example, because "lungs" are a property of "animal," every concept below that node (like "bird" or "robin") inherits that property.

    • Distance Effect: The farther the distance between two nodes in the hierarchy, the longer the cognitive processing time required to verify the relationship.

  • Limitations of the Hierarchical Model:

    • Non-Hierarchical Relationships: Semantic activation often occurs between concepts with non-hierarchical ties, such as thematic associates.

    • Unaccounted Variables: The model fails to account for reliable predictors of response times like word frequency and typicality effects (e.g., people verify "A robin is a bird" faster than "A penguin is a bird," despite both being at the same hierarchical level).

Spreading Activation Semantic Network

  • Architecture:

    • Nodes: Represented as ovals, these denote specific concepts or words.

    • Links: Lines connecting nodes that denote the strength and type of association.

    • Activation Mechanism: Links indicate that the connected nodes have been activated together in the past. When one node is activated, the activation spreads to nearby nodes.

    • Decay: The strength of activation decreases as the distance from the original node increases. This is analogous to how temperature decreases as the distance from the sun increases (McNamara, 1992).

  • The "Flat" Network Advantage:

    • Unlike the vertical nested structure of the hierarchical model, this is an interconnected web.

    • It effectively explains non-hierarchical relationships, such as thematic associates.

    • It provides a better account for context effects in language processing.

The Deese-Roediger-McDermott (DRM) Procedure

  • The Paradigm:

    • Participants are presented with a list of words that are all semantically related to a non-presented "lure word."

    • Example List (Theme: Sleep): bed, rest, awake, tired, dream, snooze, blanket, doze, slumber, nap, yawn, drowsy, snore.

    • Test Phase: Participants are asked if they saw specific words.

      • Unrelated words (e.g., chips): Easy to reject ("no").

      • Studied words (e.g., bed, nap): Correctly identified ("yes").

      • Thematically-related lure words (e.g., sleep): Participants often mistakenly "recall" or say "yes" to these words because activation spread from the studied words to the related concept.

  • Statistical Findings (Figure 8.2):

    • Recall of actual list words often reaches 60%60\% to 80%80\%.

    • Mistaken "recall" of theme lure words is remarkably high, sometimes comparable to the recall of words actually on the list.

    • Unrelated words have a very low recall/recognition rate (<10%<10\%).

Inference in Language Comprehension

  • Definition: Going beyond the literal meaning of a text; the derivation of additional knowledge from facts already known.

  • Purpose: To maintain coherence, which is the process of forming a semantically integrated whole across sentences. This includes maintaining consistency in topic, time, location, and causal relationships.

  • The Three Main Types of Inference:

    1. Logical: Follows directly from the meaning of words (e.g., hearing "bachelor" leads to the inference that the person is male).

    2. Bridging: A "backwards" inference used to connect new information to previous information in the text.

    3. Elaborative: Utilizing world knowledge to extend or embellish what is explicitly stated in the text.

  • Timing of Inference (Singer, 1994):

    • Experiment: Participants verified "A dentist pulled a tooth" after three conditions:

      1. Explicit: "The dentist pulled the tooth painlessly. The patient liked the method."

      2. Bridging: "The tooth was pulled painlessly. The dentist used a new method."

      3. Elaborative: "The tooth was pulled painlessly. The patient liked the new method."

    • Results: Bridging inferences are drawn automatically during comprehension (response times similar to explicit statements). Elaborative inferences are roughly 250 ms250\,ms slower, as they are often made later during recall.

  • Elaborative Inference and Cued Recall (Corbett & Dosher, 1978):

    • Sentences memorized:

      • High-probability: "The athlete cut out an article with scissors…"

      • Implicit: "The athlete cut out an article…"

      • Low-probability: "The athlete cut out an article with a razorblade…"

    • Finding: "Scissors" was an equally effective cue for both the high-probability and the implicit sentence. This suggests the inference wasn't necessarily drawn at encoding but was used to work backward at recall (Cue →\rightarrow Action →\rightarrow Sentence).

Practical Implications: Eyewitness Memory and the Constructive Nature of Memory

  • The Malleability of Memory:

    • Memory is not an exact replay; it is a constructive process of active reconstruction based on fragments of information, beliefs, and context.

    • Linguistic inference can "contaminate" or alter recollection.

  • Key Studies by Loftus & Palmer (1974) and Loftus & Zanni (1975):

    • Verb Choice: Participants viewed a car accident and were asked about speed using different verbs.

      • "Hit" vs. "Smashed": Those who heard "smashed" gave higher speed estimates and were more likely to report seeing broken glass a week later (even when none existed).

    • Determiners: "Did you see a broken headlight?" vs. "Did you see the broken headlight?"

      • The word "the" presupposes that a broken headlight exists, making participants more likely to incorrectly respond "yes."

  • Conclusion: While human memory is not entirely unreliable, it is susceptible to suggestion. Caution must be exercised to avoid leading questions in legal or investigative contexts.

Reference and Anaphora

  • Reference: The cognitive process of working out what linguistic expressions refer to.

  • Co-reference: When two linguistic expressions refer to the same entity (e.g., "Vlad" and "he" in "Vlad put the knife on the table. Then he forgot where it was.").

  • Anaphora Resolution:

    • A backward inference used to maintain coherence.

    • Anaphor: The material that cannot be identified in isolation (e.g., pronouns like "he," "it").

    • Antecedent: The referent or the expression being referred to.

    • Breadth of Anaphora: It is not limited to pronouns. It can include Noun phrases (e.g., "Vlad went to the cinema. The vampire enjoyed the movie.") or Verbs (e.g., "Vlad loves Boris and so does Dirk.").

  • Coping with Anaphoric Ambiguity:

    • Ambiguity occurs when there are multiple possible antecedents (e.g., "Vlad stuck a dagger in the corpse. It was made out of silver.").

    • Resolving ambiguity relies on:

      1. Guiding an explicit search.

      2. Excluding items from the search set.

      3. Avoiding explicit search altogether based on word meaning or sentence role.

Factors Affecting Anaphora Resolution

  • Parallel Function (Sheldon, 1974):

    • Comprehenders prefer to match anaphors to antecedents in the same grammatical/relevant position.

    • "Vlad sold Dirk his broomstick because he hated it." (Parallel interpretation: He = Vlad).

    • Unparallel structures cause comprehension difficulties.

  • Order-of-Mention and Gender (Arnold et al., 2000):

    • Order-of-Mention: The character mentioned first in a sentence is more accessible. Resolving a pronoun to the first-mentioned character can happen as quickly as 200 ms200\,ms.

    • Gender: Gender serves as a powerful cue for resolution when discourse cues are ambiguous (e.g., "Donald is bringing some mail to Minnie… She's carrying an umbrella").

    • Both accessibility and gender cues affect the initial process of pronoun resolution.

  • Verb Implicit Causality (Grober et al., 1978; McKoon et al., 1993):

    • Different verbs carry different implications about who is responsible for an action.

    • The "initiator" of the verb's action is usually more accessible than the "reactor."

    • NP1 Verbs (Subject Biased): "Vlad sold his broomstick to Boris because he…"

    • NP2 Verbs (Object Biased): "Vlad blamed Boris because he…" (The inference is usually that Boris did something to be blamed).

Questions & Discussion

  • Question 1: How does spreading activation explain the pattern that "a robin has wings" is judged faster than "a robin has lungs"?

    • Response: A. "wings" is closer to "robin" in the network than "lungs."

  • Question 2: Refer to the sentence "Vlad blamed Boris because he broke the vase." Parallel function predicts the interpretation of "he" to be ____; Verb implicit causality predicts the interpretation of "he" to be ____.

    • Response: D. Parallel function predicts Vlad (Subject matches Subject); Verb implicit causality predicts Boris (Boris is the likely cause of the blaming).