TL;DR
This paper introduces a novel "move-eager" left-corner parser for Minimalist Grammars that computationally models the "active-filler" strategy observed in human language processing. By formalizing how parsers handle long-distance dependencies like filler-gap constructions, this work bridges the gap between psycholinguistic theories and formal grammar parsing.
What Problem Does This Paper Address?
Human language comprehension often involves processing complex syntactic structures, particularly those with "filler-gap" dependencies. These occur when a word (the filler, e.g., a "wh"-phrase like "What") appears far from its expected grammatical position (the gap). Psycholinguistic evidence suggests that humans actively hypothesize the presence of these gaps early in processing, even before all the necessary grammatical information is available – a strategy known as "active-filler" or "active gap-finding."
The challenge has been to connect these psycholinguistic findings with formal models of parsing, especially those based on contemporary theories like Minimalist Grammars (MGs). Previous attempts to model MG parsing often focused on local dependencies or required a full derivation before processing long-distance elements, making it difficult to capture the "active" nature of human gap-finding. This paper aims to bridge this gap by developing a parser that can explicitly model this active strategy within the framework of MGs.
Methods
The core contribution is the development of a move-eager left-corner Minimalist Grammar parser. This approach builds upon existing left-corner parsing techniques, which process input from left to right, but with a crucial modification: it incorporates "move-eager" transitions.
Here's a breakdown of the key methodological aspects:
- Minimalist Grammars (MGs): The parser is grounded in the formal framework of MGs, which provide an explicit computational theory of syntax derived from Chomsky's work.
- Left-Corner Parsing: This strategy processes input incrementally, building structures from left to right. It's considered psychologically plausible due to its ability to handle memory demands in embedding structures.
- Move-Eager Transitions: This is the novel aspect. Unlike previous MG parsers that might delay processing movement dependencies, this parser actively integrates movement information as soon as possible. Specifically, it establishes the connection between a filler (e.g., "What") and its potential gap site earlier in the parsing process, even before the entire clause structure is resolved. This is achieved through specific parsing operations (C0, C1, C2, C3) that connect newly derived structures with stored implications, allowing for the "eager" establishment of movement-related dependencies.
- Search Space Formalization: The parser's operations define a search space where choices correspond to different parsing strategies. The paper argues that the preference for "active gap-finding" in humans can be modeled as a preference for certain transitions (like C3) over others (like C2) within this search space.
Datasets
Not discussed in the paper. The paper focuses on theoretical modeling and uses example sentences for illustration rather than empirical data from large corpora or experimental datasets.
Key Insights and Ideas
- Computational Model for Active Gap-Finding: The paper provides the first formal, computational model that can explicitly represent and explore the "active-filler" strategy within the Minimalist Grammar framework.
- Bridging Psycholinguistics and Formal Grammar: It demonstrates how formal parsing mechanisms can be designed to directly mirror psycholinguistic observations about human sentence processing, moving beyond abstract metrics to algorithmic-level questions.
- Move-Eagerness as a Key to Active Processing: The "move-eager" aspect of the parser is crucial. It allows the parser to establish filler-gap connections earlier, reflecting the human tendency to hypothesize gaps proactively.
- Search Space as a Decision Arena: The parser's search space is framed as the locus of decision-making for parsing strategies. The preference for active gap-finding can be seen as a preference for specific paths within this space, analogous to how "Late Closure" is modeled in other parsing contexts.
- Explaining the "Filled-Gap Effect": The model provides a framework for understanding phenomena like the "filled-gap effect" (processing slowdown when a potential gap is unexpectedly filled), suggesting it arises from backtracking when an active-filler hypothesis is disconfirmed.
Key Results
The paper doesn't present quantitative results in terms of accuracy or speedup on specific datasets, as it's a theoretical and algorithmic development. However, the key results are:
- A formal, sound, and complete parser for MGs that incorporates move-eager transitions.
- Demonstration of how the parser's search space can represent the choice between active gap-finding and other strategies.
- Illustrative traces (Figures 5 & 6) showing how the parser processes sentences with filler-gap dependencies, highlighting the "eager" establishment of these connections. For example, in the sentence "What John buys," the parser posits the gap site before shifting the verb "buys" (Step 6 in Figure 5).
Limitations and Future Work
- Formalization of the "Active-Filler" Preference: While the parser's search space allows for modeling the active-filler strategy, the paper acknowledges that it doesn't explain why humans prefer this strategy. The model includes a stipulation for this preference, rather than deriving it from more fundamental principles.
- Island Constraints: The paper notes that a key property of active gap-finding is its sensitivity to "islands" (syntactic structures that block movement). Future work aims to investigate whether this island-sensitivity can emerge as a natural consequence of the grammatical encoding within this parsing framework.
- Overt Complementizers: The paper suggests that the model might offer insights into languages with overt complementizers that co-occur with fronted wh-phrases, as the integration of the filler happens after the (potentially null) complementizer is shifted.
- "Hyper-Active" Gap-Finding: The parser's mechanism of establishing connections from a daughter node to predict its sister (rather than waiting for both to be fully resolved) aligns with findings on "hyper-active" gap-finding reported in some psycholinguistic studies.
Summary updated