TL;DR
This paper presents the first computational algorithms for determining anaphor and pronoun binding according to Chomsky's Binding Theory (BT). By formalizing BT principles and their interactions with linguistic structures like argument chains, the authors provide a foundation for computational linguistics tools that can verify grammatical correctness related to referential dependencies.
What Problem Does This Paper Address?
Linguistic theories often describe how pronouns and anaphors (like "himself" or "him") must relate to other noun phrases (NPs) in a sentence to be grammatically correct. This is known as Binding Theory (BT). For example, an anaphor like "himself" needs an "antecedent" (the NP it refers to) within the same sentence, while a pronoun like "him" can either refer to an antecedent or to something outside the sentence. The challenge is to translate these abstract linguistic rules into concrete computational procedures that can verify these properties for any given sentence. Previous computational approaches often focused on outputting actual references, which the authors argue is insufficient. This paper aims to provide algorithms that compute the binding requirements themselves, focusing on the constraints imposed by BT.
Methods
The authors propose two algorithms, Algorithm 1 for Principle A (anaphors) and Algorithm 2 for Principle B (pronouns) of Binding Theory. These algorithms operate on a parse tree representation of a sentence. They utilize concepts like:
- C-command: A structural relationship between nodes in a parse tree, crucial for defining binding domains.
- Complete Functional Complex (CFC): A specific structural domain within a parse tree that defines the local environment for binding.
- Agreement: A feature-matching mechanism between NPs.
- Argumental Chains: Structures representing movement or co-reference, particularly important for handling pro-drop languages (languages where subjects can be omitted).
The algorithms systematically traverse the parse tree, checking for potential antecedents (for anaphors) or disjoint referents (for pronouns) within their respective local domains (CFCs). They also incorporate mechanisms to handle circularity and interactions with argumental chains, especially in pro-drop contexts.
Datasets
Not discussed in the paper. The paper focuses on theoretical algorithms and does not present experimental evaluation on specific datasets.
Key Insights and Ideas
- Focus on Constraints, Not Just References: The paper argues that BT's primary role is to constrain possible referential assignments, not to directly output them. Their algorithms reflect this by identifying valid antecedents or disjoint referents based on BT rules.
- Formalizing Local Domains (CFCs): The precise definition and computational handling of Complete Functional Complexes are crucial for correctly applying BT principles. The paper emphasizes the dynamic nature of these domains, influenced by structural and lexical interactions.
- Handling Pro-Drop Languages: The algorithms are designed to account for languages like Italian where subjects can be omitted (pro-drop). This requires careful management of empty categories and their relation to other NPs, particularly concerning the interaction of binding principles with argumental chains.
- Computational Complexity is Tractable: The authors demonstrate that their algorithms for recognizing Principle A and B violations run in polynomial time, suggesting that computational verification of these core linguistic constraints is feasible.
Key Results
- The paper presents two algorithms (Algorithm 1 for Principle A, Algorithm 2 for Principle B) that computationally verify the constraints imposed by Binding Theory on anaphors and pronouns.
- The algorithms are shown to correctly handle interactions with argumental chains and pro-drop phenomena.
- The running time of both algorithms is proven to be linear in the length of the sentence (O(n)), indicating an efficient computational approach.
Limitations and Future Work
- Scope of BT: The paper explicitly states that Principles A and B are not sufficient on their own to provide actual referential assignments. They only restrict the search space.
- Split Antecedents: The paper acknowledges that handling "split antecedents" (where a pronoun refers to a set of NPs) is complex and could lead to exponentially large search spaces if not managed carefully. This is identified as a challenge for future work.
- Principle C and Other Phenomena: The current work focuses on Principles A and B. The authors mention that their broader project extends to Principle C and weak-crossover cases, suggesting these are areas for further computational development.
- Empirical Validation: While the algorithms are theoretically sound, the paper does not include empirical validation on natural language corpora. Future work would likely involve implementing and testing these algorithms on real-world linguistic data.
Summary updated