lecture 7 - sentence processing Flashcards

Question

What are minimal attachment and late closure?

Answer 1

Strategies the processor follows when forming the syntactic tree

Answer 2

* Initial parsing preferences * Later information contradicts initial preference Garden path sentences

Answer 3

(Rayner, Carlson & Frazier, 1983) Attachment - Minimal The spy saw the cop with binoculars but the cop didn’t see him. Non-minimal The spy saw the cop with the revolver but the cop didn’t see him. If you followed minimal attachment to follow the second sentence, you’d interpret it to mean that the spy saw the cop using a revolver to see through. So then you’d have to backtrack and undo your initial interpretation. Whereas you wouldn’t if you understood the first sentence using minimal attachment. Rayner, Carlson & Frazier, 1983 Eye tracking As predicted by minimal attachment Present people with 2 sorts of sentences – one with crrect meaning obeying minimal attachment and one not People in eye tracking – non minimal attachment had more backtracking to earlier parts of sentence

Answer 4

attachment - minimal backtrack - no The spy saw the cop with the binoculars ... attachment - Non- minimal backtrack - Yes The spy saw the cop with the revolver ...

Answer 5

Attachment - minimal The couple admired the house with a friend but knew that it was over-priced. non-minimal The couple admired the house with a garden but knew that it was over-priced. T&M argues the semantic expectations lead you to expect some information about what the spy used to see with. Semantic context leads to the expectation that some information about the house will be given. In the first case, the sentence follows minimal attachment but violates the semantic expectation. In the second, the sentence violates minimal attatchment but follows semantics. Taraban & McClelland, 1988 Similar structure to previous examples Frazier’s parsing strategy predicts that minimal attachment parses will be considered first and therefore should be faster. For some sentences those predictions are right, but Taraban & McClelland showed that the predictions are wrong for other sentences. Semantics favour info about house So, it doesn’t look like syntax finishes one parse, gives it to semantics, and then tries another one next if necessary. Semantics seems to interact with syntactic parsing more than that. There seem to be two ways to explain this.

Answer 6

The couple admired the house with a friend ... attachment - Minimal backtrack - Yes The couple admired the house with a garden ... attachment - Non-minimal backtrack - No So, it doesn’t look like syntax finishes one parse, gives it to semantics, and then tries another one next if necessary. Semantics seems to interact with syntactic parsing more than that. There seem to be two ways to explain this. Couple with friend http://tbn0.google.com/images?q=tbn:F_EUW52gAjmdMM:http://bp1.blogger.com/_itKXWJIFGGU/RoKl0wueOwI/AAAAAAAAAY8/twvtPkLGFgg/s320/ashton-holding-hands-with-demi-and-penelope.jpg 6 Nov 2008

Answer 7

* Amazing question: how do we parse sentences when there is ambiguity? Not clear whether LC or MA is the answer, but certainly a lot of data to explain.

Answer 8

* Immediacy * Lexical ambiguity Syntactic ambiguity

Answer 9

Some psycholinguists (e.g., Frazier, 1989; Swinney, 1979), following Chomsky (1981b), argue that initial lexical access and syntactic parsing are autonomous, unaffected by prior semantic information or background knowledge. Other researchers (e.g., Taraban & McClelland, 1988; Trueswell et al., 1994), aligning with challenges to Chomsky (e.g., Langacker, 1988), claim that semantic information guides lexical access and syntactic analysis, viewing processing at lexical, syntactic, and semantic levels as highly interactive. The modular versus interactive debate centers on differences in information flow. A related question concerns the coordination of semantic and pragmatic factors. For instance, when reading a metaphor, literal processing might precede pragmatic interpretation, or pragmatic factors might influence semantic analysis early on, enabling simultaneous literal and metaphoric comprehension (Gibbs, 1994; Glucksberg, 1991). Sentence comprehension appears to proceed in stages—from lexical access to pragmatic analysis—with earlier stages working independently. Three possibilities arise: Each stage is distinct and autonomous; Stages are fully interactive, with semantic and pragmatic information available throughout; A mixed model, where early stages like lexical access are autonomous but later ones are interactive. Investigating these processes is challenging because comprehension is rapid and largely automatic. To explore these mental events, we must first study methods for investigating sentence processing, followed by a review of data on modular versus interactive processing.

Answer 10

Concise Summary: Sentence Comprehension and Ambiguity in Psycholinguistics Comprehension and Memory: People often recall the meaning or situation described by a sentence rather than its exact wording. This suggests world knowledge plays a strong role in sentence comprehension. Influence of Prior Knowledge: Studies like Fillenbaum's show that people paraphrase anomalous sentences into more logical forms without realizing it, indicating that understanding is shaped by expectations and real-world knowledge. Immediacy of Interpretation: Sentence processing happens word-by-word, not after receiving the full sentence. This helps manage working memory constraints but can lead to misinterpretation (e.g., garden-path sentences). Ambiguity and Parsing: Lexical ambiguity (e.g., “bug”) and syntactic ambiguity (e.g., “warned” in different contexts) are common and informative for studying comprehension. Sentence processing involves quickly integrating words semantically and syntactically, but not always completely before moving on. Modular vs. Interactive Models: Modular view: Initial processing is syntactic and independent of context; semantic info comes later. Interactive view: Context can influence early interpretation, even affecting which word meanings are activated. Research Methods: Techniques like semantic priming and eye-tracking (e.g., fixation durations) help reveal real-time comprehension processes, especially in ambiguous or garden-path sentences. Ecological Validity: While some criticize ambiguity-focused research as artificial, ambiguity is actually common in natural language, making such studies relevant to understanding everyday sentence comprehension.

Answer 11

he reason that many of the studies of sentence processing make use of one form of ambiguity or another is because the major views of sentence processing clash most sharply when ambiguity is considered. On the one hand, a strictly modular view assumes that separate processors analyze lexical, syntactic, and semantic information, and that semantic information cannot influence our initial lexical or syntactic analysis even if a word or a syntactic structure is ambiguous. On the other hand, an interactive view sees the resolution of both lexical and syntactic ambiguity as being a matter of constraints based on simultaneous consideration of lexical, syntactic, and semantic information.

Answer 12

Swinney (1979) studied how people process lexical ambiguity using sentences containing ambiguous words like bug. Participants heard sentences and were shown a visual target word for a lexical decision task immediately or shortly after hearing the ambiguous word. Results showed both meanings (e.g., insect and spy) were briefly activated, but only the contextually appropriate meaning remained active after a short delay. This supports the modular view: initial lexical access is context-independent, and context influences meaning afterward. However, later studies found context can influence initial access, depending on: Meaning dominance – how frequent each meaning is in general language use. Strength of context – how strongly the sentence suggests a particular meaning. For instance: In sentence (8) "I opened a checking account at the bank", only the dominant financial meaning of bank is activated. In sentence (9) "The fisherman waited by the bank", both meanings may be activated briefly due to weaker support for the subordinate (riverbank) meaning. Simpson, Tabossi, and others showed that selective access occurs when a strong context supports the dominant meaning, while multiple access occurs when context is weak or the subordinate meaning is intended. Duffy et al. (1988) used eye-tracking to measure reading times for ambiguous words. They manipulated: Whether the word was equibiased (two equally common meanings) or nonequibiased (dominant and subordinate meanings). Whether disambiguating context came before or after the ambiguous word. Findings: When context came before the ambiguous word: Nonequibiased words (with subordinate meanings intended) showed longer reading times, reflecting multiple access. Equibiased words showed no slowdown, suggesting selective access helped by context. When context came after the ambiguous word: Equibiased words were read more slowly than unambiguous words, indicating both meanings were accessed. Nonequibiased words showed no initial slowdown, but later context (favoring the subordinate meaning) caused slower reading in the resolution region. Conclusion: Lexical access is influenced by both bottom-up factors (meaning dominance) and top-down context, but context effects are limited. The findings support a hybrid view: early processing is generally modular, but strong context can guide meaning selection under certain conditions.

Answer 13

Modular View of Parsing Rayner and colleagues (Frazier & Rayner, 1982, 1987) proposed a modular view, where syntactic parsing operates independently of semantic processing—at least initially. When a sentence is ambiguous, the parser commits to one syntactic interpretation without immediate semantic input unless comprehension fails. Parsing Strategies Late Closure Definition: New elements are attached to the most recent clause or phrase. Example: In “Martin had a date with the sister of the teacher who was named…”, “who was named” is usually attached to “the teacher.” Evidence: Eye-tracking studies (Frazier & Rayner, 1982) showed that readers spend more time parsing garden-path sentences like: (12) “Since Jay always jogs a mile seems like a short distance…” Misinterpretation due to assuming “a mile” is the object of “jogs.” Cross-linguistic Evidence: English speakers tend to use late closure, but Spanish speakers often attach clauses to earlier noun phrases. Minimal Attachment Definition: Prefer syntactic structures with the fewest nodes (i.e., simplest). Example: In “The spy saw the cop with binoculars…”, the default interpretation is that the spy had the binoculars. Evidence: Eye movement data (Rayner et al., 1983) shows delayed comprehension when minimal attachment leads to incorrect interpretation (e.g., when "with a revolver" creates an implausible reading). Challenges: Semantic context may drive interpretation more than syntax (Taraban & McClelland, 1988). Role of Semantics Evidence against strict modularity: Readers use semantic context to adjust parsing decisions. In sentences like (15–16) and (17–20), comprehension difficulty varies depending on the plausibility of the subject performing the verb (e.g., inanimate nouns like “evidence” influence parsing). Trueswell et al. (1994): Garden-path effects are stronger when semantic expectations are violated. Interactive vs. Modular Debate Some models (e.g., “syntax proposes, semantics disposes”) allow for a syntactic module that generates structures, which semantics then quickly evaluates (Altmann & Steedman, 1988). The faster nonsyntactic information is used, the harder it is to uphold a strictly modular view.

Answer 14

Recent research has expanded beyond written language to explore spoken language processing, offering new insights into whether syntactic parsing is modular or influenced by broader context. Tanenhaus and colleagues developed an eye-tracking method to monitor how listeners process spoken instructions in real-time visual contexts. For example, in the ambiguous sentence "Put the saltshaker on the envelope in the bowl," listeners used visual cues to disambiguate syntactic structure before minimal attachment could lead them astray. Eye movements showed listeners quickly looked at the correct referents, indicating they used visual context early to influence syntactic parsing. This evidence challenges strictly modular views by showing that nonsyntactic information (like visual context) can influence initial syntactic decisions. A follow-up experiment confirmed that when visual context didn't clarify the referent, listeners initially misparsed the sentence, supporting the idea that parsing is sensitive to available cues. In addition to visual input, prosody (intonation and pauses) also helps listeners parse sentences correctly. For instance, a pause in the sentence "After Bill left, Tamara returned home" helps avoid a garden-path misinterpretation. Overall, these findings suggest that sentence processing is flexible and interactive, rapidly integrating visual, prosodic, syntactic, and semantic information, rather than being strictly modular or bottom-up.