TL;DR
This paper argues that improving the accuracy of the syntax-semantics interface in Head-Driven Phrase Structure Grammar (HPSG) requires moving beyond linear ordering of grammatical arguments. By introducing a "branching" (non-linear) obliqueness hierarchy for arguments, the authors demonstrate how to elegantly solve previously intractable binding puzzles in languages like Toba Batak, Japanese, Chinese, and Portuguese.
What Problem Does This Paper Address?
In computational linguistics, particularly within frameworks like HPSG, the "SUBCAT" (subcategorization) feature plays a crucial role in linking syntax and semantics. Traditionally, this feature encoded information about a verb's arguments, including their order. This linear order was used for various linguistic generalizations, such as word order and binding. However, recent research has shown that a single linear ordering is insufficient to capture all these generalizations. Specifically, the rules governing how arguments bind to each other (like reflexives) don't always align with the order in which arguments appear on the surface. This paper tackles the problem of how to accurately represent argument relationships to improve the syntax-semantics interface, especially when dealing with complex binding phenomena that defy simple linear ordering.
Methods
The paper's core contribution is the proposal of a branching obliqueness hierarchy for arguments, specifically for the newly introduced ARG-S (argument structure) feature. This ARG-S feature is a separate list from the SUBCAT list, which retains the linear surface order.
The authors build upon the work of Manning and Sag (1995), who first proposed splitting the SUBCAT information into two lists: one for surface concatenation (valence) and one for binding relations (argument). This paper takes it a step further by arguing that the ordering within the ARG-S list should not be strictly linear but can be branching.
The key methodological innovation is the redefinition of o-command (a crucial concept for binding principles) to accommodate these non-linear, branching hierarchies. This allows for cases where arguments are considered equally oblique, or where the traditional linear precedence no longer dictates binding relationships.
Datasets
The paper does not introduce new datasets but rather uses existing linguistic examples from various languages to illustrate its points. These include:
- Toba Batak: Examples of reflexive constructions in objective and active voice.
- Japanese: Reflexives in causative constructions.
- Chinese: Subject-oriented reflexive
ziji. - Russian: Reflexives in passive constructions.
- Portuguese: Reflexives in oblique complements.
Key Insights and Ideas
- Linearity is Limiting: The paper strongly argues that a single linear ordering for arguments is fundamentally inadequate for capturing the complexities of binding phenomena across languages.
- Branching is Better: Introducing a non-linear, branching hierarchy for argument obliqueness (within the
ARG-Sfeature) provides a more flexible and accurate way to model these relationships. - O-command Needs Rethinking: The traditional definition of o-command, based on linear precedence, needs to be extended to handle branching hierarchies, particularly to define cases of equal obliqueness.
- Unified Solution: The proposed branching obliqueness model offers a unified and elegant solution to a range of previously disparate and difficult binding puzzles across diverse languages.
Key Results
The paper presents a theoretical framework that, when applied to the discussed linguistic examples, correctly predicts the grammaticality of sentences involving reflexives. For instance:
- It correctly accounts for the binding of reflexives in Toba Batak, where linear order alone failed.
- It explains the behavior of Japanese reflexives in causative constructions, where subject-orientation is key.
- It resolves the puzzle of Chinese
zijibeing subject-oriented. - It clarifies the behavior of Russian reflexives in passive constructions, where lexical rules alter obliqueness.
- It solves problematic reflexive binding in Portuguese oblique complements.
The paper demonstrates that by allowing ARG-S values to have a branching configuration, the empirical adequacy of Binding Theory and the accuracy of the syntax-semantics interface are significantly improved.
Limitations and Future Work
- Limited Scope of O-command: The paper acknowledges that the definition of (non-linear) o-command provided is a partial specification, and further cases where elements are not equally oblique but also don't precede each other were not fully discussed due to space constraints. The authors suggest that the proposed definition can be extended to cover these cases.
- Empirical Confirmation: While the paper provides strong empirical motivation, it suggests that further empirical confirmation for the need for branching configurations in o-command definitions would be beneficial, particularly in cases where lexical rules change obliqueness relations in ways that necessitate such configurations.
- Language-Specific Principles: The paper touches upon the idea that different languages might require different configurations of the
ARG-Svalue (e.g., linear vs. branching), hinting at future work in exploring the typology of these argument structures.
Summary updated