Control Verbs, Argument Cluster Coordination and MCTAG

Djamé Seddah, Benoît Sagot, Laurence Danlos2010

TL;DR

This paper extends the Multi-Component Tree Adjoining Grammar (MCTAG) framework to handle complex linguistic phenomena like non-local ellipsis and argument cluster coordination. By introducing "ghost trees" and a novel interpretation of derivation trees, the authors provide a more robust and unified analysis for these challenging syntactic structures.

What Problem Does This Paper Address?

Linguistic theories often struggle to elegantly account for phenomena involving ellipsis, where parts of a sentence are omitted but understood. Specifically, this paper tackles two challenging areas:

  1. Non-local Elliptic Coordinations: These occur when elided elements are not adjacent to their antecedents, often found in constructions with control verbs (e.g., "John hopes to sleep," where "John" is the subject for both "hopes" and "to sleep"). Existing models, particularly Tree-Local MCTAGs (TL-MCTAGs), are limited to local dependencies and cannot handle these non-local relationships effectively.
  2. Argument Cluster Coordination (ACC): This involves coordinating multiple arguments, often with ellipsis, leading to complex sentence structures (e.g., "John gives Mary a flower and Paul a ring"). While some approaches exist, a unified framework that can handle both ACC and other forms of ellipsis remains a challenge.

The authors aim to bridge this gap by extending MCTAGs to provide a more comprehensive and unified analysis of these phenomena.

Methods

The paper proposes an extension to the Multi-Component Tree Adjoining Grammar (MCTAG) framework, building upon previous work on MCTAGs with Local Shared Derivation (MCTAG-LSD). The key innovations include:

  • Ghost Trees: For elliptical constructions, a "ghost tree" is introduced. This is essentially a copy of a lexicalized tree but anchored by an empty element. The presence of its corresponding ghost tree within the same MC-Set allows the system to retrieve the missing information.
  • Control Verb Analysis: Control verbs are modeled using "ghost trees" and a mechanism that allows for the transfer of controlled arguments. This is achieved by associating a control tree with a single node that shares a derivation with the controlled node.
  • Novel Derivation Tree Interpretation: The authors propose an interpretation of MCTAG derivation trees that mixes the views of Kallmeyer (2005) and Weir (1988). This hybrid approach is crucial for providing an adequate derivation graph for argument cluster coordinations.
  • Extension to Non-Local Ellipsis: The core MCTAG-LSD framework is extended to handle non-local elliptic coordinations. This involves allowing derivations to be shared across different tree sets, moving beyond the limitations of Tree-Local MCTAGs.

Datasets

Not discussed in the paper. The paper focuses on theoretical linguistic analysis and proposes a formalism rather than empirical evaluation on specific datasets.

Key Insights and Ideas

  • Ghost trees are a powerful tool for handling ellipsis: By representing elided elements with "ghost trees" within the same MC-Set, the framework can elegantly recover missing information without complex ad-hoc mechanisms.
  • Control verbs can be unified with ellipsis analysis: The proposed model effectively integrates the analysis of control verbs with the general mechanism for handling ellipsis, demonstrating a more cohesive linguistic theory.
  • A mixed interpretation of derivation trees is key for ACC: The authors highlight that a novel interpretation of MCTAG derivation trees, combining existing approaches, is essential for correctly analyzing argument cluster coordination.
  • Moving beyond local dependencies is crucial for non-local ellipsis: The extension to handle non-local ellipsis signifies a significant step in the expressiveness of MCTAGs, allowing for more complex and realistic linguistic phenomena to be modeled.

Key Results

The paper presents a theoretical framework that:

  • Provides an analysis for non-local elliptic coordinations, particularly those involving control verbs.
  • Offers a unified approach to argument cluster coordination by leveraging the proposed extension.
  • Demonstrates that the extended MCTAG framework (specifically, Non-Local MCTAGs) is necessary to cope with unbounded ellipsis, while Set Local MCTAGs are sufficient for Argument Cluster Coordination.
  • Shows that the proposed framework, while extending the expressive power of MCTAGs, maintains the weak generative capacity of MCTAGs.

Limitations and Future Work

  • Ambiguity in Ghost Trees vs. Shared Derivations: The paper acknowledges a potential source of ambiguity when ghost trees can substitute for shared derivations. They suggest that precedence rules could circumvent this, but it also highlights the precise ambiguity needed for ACC analysis.
  • Type System for ACC: For a proper dependency structure in argument cluster coordination, the authors suggest that a mechanism for determining whether an MC-Set denotes a unique event/individual or a list of instances might require more than just syntactic information, potentially relying on an adequate type system.
  • Scope Information: The authors note that distinguishing between narrow and wide scope readings in complex coordination scenarios would require a richer model that can infer scope information from all trees within an MC-set. They suggest integrating models like Kallmeyer and Joshi (2003) for this purpose.
  • Future Directions: The paper opens avenues for further research into the generative power needed for appropriate derivation structures and the precise boundary of syntactic analysis. It also points towards integrating richer semantic information, such as scope, into the framework.

Summary updated

Abstract

In this paper1 we present an extension of MCTAGs with Local Shared Derivation (Seddah, 2008) which can handle non local elliptic coordinations. Based on a model for control verbs that makes use of so-called ghost trees, we show how this extension leads to an analysis of argument cluster coordinations that provides an adequate derivation graph. This is made possible by an original interpretation of the MCTAG derivation tree mixing the views of Kallmeyer (2005) and Weir (1988). 1
BibTeX
@misc{seddah2010control,
  author = {Djamé Seddah and Benoît Sagot and Laurence Danlos},
  title = {{Control Verbs, Argument Cluster Coordination and MCTAG}},
  year = {2010},
}

Loading related content...

Figure