Lecture 4: Sentence Comprehension

Sentence Comprehension

• Sentence comprehension refers to the syntactic/structural analysis and interpretation of sentences.
• Sentence parsing focuses on syntactic/structural analysis within sentence comprehension.
• Sentence processing is a broader term that includes both sentence comprehension and sentence production.

Ambiguity in Sentence Processing

• There are two types of ambiguous sentences:
  • Temporarily ambiguous sentences: ambiguity occurs at some point in the sentence but is resolved later by syntactic or semantic/pragmatic constraints.
  • Globally ambiguous sentences: even after reading the whole sentence, there are multiple ways to analyze the structure, or the syntactic ambiguity is not resolved.
• Studying ambiguity in sentence processing helps us understand what information is used when grammar doesn't lead to a specific analysis.

Garden Path Model

• The garden path model attempts to explain parsing (syntactic analysis) of both ambiguous and unambiguous sentences.
• The model assumes incremental processing, where words are immediately incorporated into previous structures from left to right.
• The model follows two principles:
  • Minimal attachment: attach the ambiguous phrase to the structure using the fewest possible additional nodes.
  • Late closure: attach the incoming phrase to the phrase currently being processed, making it as low in the tree structure as possible.

Experimental Evidence for Minimal Attachment

• Rayner et al. (1983) used eye tracking to measure participants' reading times and focus on ambiguous sentences.
• The study supported the minimal attachment principle, as participants focused more on the second half of the sentence and reading time was longer in the noun phrase condition.

Late Closure

• Late closure is used when minimal attachment does not resolve ambiguity.
• It involves attaching the incoming phrase to the phrase currently being processed, making it as low in the tree structure as possible.
• Traxler et al. (1998) used eye tracking to measure participants' reading times on high and low attachment sentences.
• The study supported the late closure principle, as participants showed higher reading times for the last phrase in the high attachment condition.

Garden Path Model Assumptions

• The garden path model assumes:
  • Syntactic information is used first.
  • The model is modular (initial syntactic processing is not influenced by non-syntactic information).
  • The model is serial (only a single analysis is adopted at a time).
  • The model is a reanalysis model (difficulty occurs when the initial analysis conflicts with later information).

Constraint-Based Models

• Constraint-based models claim that syntactic processing is not modular and non-structural information is used immediately.
• They are interactive models, parallel models, and competition models.
• The three properties of constraint-based models are:
  • Interactive models: all sources of information are used immediately and interact with each other.
  • Parallel models: all analyses of ambiguous structure are activated in parallel.
  • Competition models: difficulty occurs when two or more analyses are equally activated by different sources of information.

Trueswell et al. (1994) Study

• The study used eye tracking to measure reading times on sentences with animacy and ambiguity.
• The results supported constraint-based models, as reading times were highest in the animate-ambiguous condition and lowest in any inanimate condition.

Critique and Repeat of Trueswell et al. (1994) Study

• Clifton et al. (2003) suggested that the original study's results were a null effect and could not be interpreted.
• Clifton et al. (2003) repeated the study and found that reading times were different in the two inanimate conditions, supporting the garden path model.