TL;DR
This paper proposes a novel way to handle constituent coordination within the Lexical-Functional Grammar (LFG) framework. By treating coordinate structures as flat, unstructured sets of functional elements, LFG can uniformly distribute properties across coordinated elements without needing special rules for each type of coordination. This elegantly solves long-standing issues in linguistic theory and computational linguistics regarding how to represent and process coordinated phrases.
What Problem Does This Paper Address?
Coordinating phrases like "John bought apples and John ate apples" or "John bought and ate apples" has been a persistent challenge for linguistic theories. Traditional approaches, like the Coordinate Reduction rule, struggled with the complex restrictions on what can be coordinated and how meaning is preserved. Computational systems often resort to ad-hoc parsing mechanisms. A key difficulty is explaining why certain coordinations are grammatical while others are not, even when the individual components seem fine (e.g., "The girl promised and persuaded John to go" is ungrammatical). Furthermore, capturing the subtle semantic differences between coordinations (e.g., whether one or multiple agents are involved) has been problematic. This paper aims to provide a unified and principled account of constituent coordination within the LFG framework, integrating it with existing grammatical principles.
Methods
The core innovation lies in extending the LFG formalism to handle coordination using flat, unstructured sets of f-structures. Instead of treating a coordinated phrase as a single, complex functional structure, it's represented as a set where each element is the f-structure of one of the coordinated constituents.
Key methodological contributions include:
- Set-based Representation: Coordinate structures are modeled as sets of f-structures. This allows for an arbitrary number of elements and avoids hierarchical dominance issues.
- Generalized Function Application: The standard LFG function application mechanism
(f a) = vis extended to sets. For a set of functionss,(s a) = vholds ifvis the generalization of(f a)for allfins. This "generalization" is defined recursively and essentially means that properties asserted for the set as a whole must hold for all its elements. This is crucial for distributing shared elements (like subjects or objects) across coordinated verbs. - Functional Uncertainty Integration: The paper shows how this set-based approach can interact with functional uncertainty, a mechanism in LFG that allows for flexible functional relationships. This is important for explaining phenomena like relative clauses where the head noun can play various roles within the clause.
- Semantic Projection for Conjunctions: The identity of the conjunction (e.g., "and," "or") is not encoded in the functional structure but is handled at the semantic level via a semantic projection (
σ). This allows for syntactic generalizations to be made independently of the specific conjunction used.
Datasets
Not discussed in the paper. The paper focuses on theoretical linguistic analysis and formal grammar development, using example sentences rather than formal datasets for training or evaluation.
Key Insights and Ideas
- Sets as a Natural Fit for Coordination: The paper argues convincingly that flat, unstructured sets are the most appropriate mathematical objects for representing coordinate structures in LFG. They naturally accommodate the unbounded nature of coordination and the lack of hierarchical scope between conjuncts.
- Uniform Distribution of Properties: The generalized function application to sets elegantly explains how shared elements (subjects, objects) are distributed across coordinated verbs or other constituents. This avoids the need for ad-hoc rules for each coordination type.
- Resolving Semantic Ambiguities: By separating the functional representation of coordination from the semantic representation of the conjunction, the paper provides a mechanism to distinguish between different semantic interpretations of coordinated sentences, as seen in the "a girl saw and talked" vs. "a girl saw and a girl talked" example.
- Functional Uncertainty and Coordination Synergy: The integration of functional uncertainty with set-based coordination provides a powerful way to account for complex dependencies, such as those found in relative clauses modifying coordinated structures.
- Decoupling Syntax and Semantics for Conjunctions: Encoding conjunction identity solely at the semantic level simplifies the functional grammar and allows for broader syntactic generalizations.
Key Results
The paper presents a formal extension to Lexical-Functional Grammar that accounts for constituent coordination. While no quantitative results are provided, the key qualitative findings are:
- A unified theoretical treatment of constituent coordination within LFG.
- An explanation for the grammatical constraints on coordination by distributing properties across set elements.
- A mechanism to differentiate semantic interpretations of coordinated sentences.
- A formal integration of coordination with functional uncertainty.
Limitations and Future Work
- Non-constituent Coordination: The paper explicitly states it focuses on constituent coordination and does not address non-constituent coordination, which presents even more complex challenges.
- Complexity of Generalization: While the concept of generalization for sets is introduced, its recursive definition and computational implications might require further exploration for practical implementation.
- Semantic Projection Details: The paper outlines the semantic projection (
σ) for conjunctions but notes that a more detailed discussion of semantic structures and their interaction with functional structures is provided in other works.
Future work suggested by the authors includes exploring a wider range of coordination phenomena and providing more general linguistic motivation for the proposed approach, as detailed in Bresnan, Kaplan, and Peterson (forthcoming). The interaction with functional uncertainty also opens avenues for further research into complex dependency structures.
Summary updated