TL;DR
This paper presents a parallel parser for Government-Binding (GB) Theory, implemented in PARLOG. The goal is to explore the computational aspects of GB's modular components and their interactions, serving as a linguistic tool for investigating concurrency and autonomy within the theory.
What Problem Does This Paper Address?
Parsing, the process of analyzing a sentence's grammatical structure, has traditionally been approached sequentially. While parallel parsing has been explored, few linguistically-oriented, GB-based parallel parsers exist. Government-Binding Theory itself is a highly modular framework, making it an ideal candidate for exploring parallel processing. This paper addresses the challenge of creating a parser that not only implements GB principles but also allows for a formal examination of how its various modules (like Binding Theory, Control Theory, etc.) could operate concurrently.
Methods
The authors developed a parser using PC-PARLOG, a parallel logic programming language. The parser employs a hybrid top-down and bottom-up strategy, directly recovering licensing relations (like thematic roles and binding relationships) rather than full phrase structures. Key procedures include detect_gap (identifying implicit elements like traces or PROs), project (constructing maximal projections), and license (indexing and assigning thematic roles). The implementation simulates parallelism through time-sharing on a single processor, but the focus is on specifying concurrency at a high level of abstraction. The parser is designed to be modular, transparently encoding each GB module for explicit investigation.
Datasets
Not discussed in the paper. The paper focuses on the theoretical implementation and linguistic coverage of a subset of English explained by GB Theory, rather than empirical evaluation on specific datasets.
Key Insights and Ideas
- GB Theory as a Natural Fit for Parallelism: The modular nature of Government-Binding Theory makes it a promising framework for exploring concurrent processing.
- Focus on Licensing Relations: The parser prioritizes identifying licensing relations (like thematic roles and binding) over generating explicit phrase structures, reflecting a different perspective on parsing output.
- Transparency for Linguistic Exploration: The design emphasizes transparent encoding of GB modules, allowing linguists to explicitly study processes within and across these components.
- Deterministic Parsing with Delayed Actions: The parser is deterministic and capable of delaying actions like node projection and lexical disambiguation until more information is available, potentially mirroring cognitive processes.
- Potential for Porting to Parallel Hardware: The PARLOG implementation is designed to be easily portable to actual parallel logic machines as they become more accessible.
Key Results
The paper demonstrates the parser's ability to detect implicit elements (PRO, trace) and their binding relationships, as illustrated by the example sentence "John was persuaded to leave." The output includes identified licensing relations and co-indexed elements. The authors suggest that the system provides a foundation for future research into the cognitive plausibility of parallel parsers and offers insights into parallel parsing and human sentence processing.
Limitations and Future Work
The authors acknowledge that the current implementation simulates parallelism and runs on a single processor, lacking true parallelism. Derivational morphology is minimal in the current version. Future directions include expanding the parser's coverage within the GB framework to include adjuncts and investigating the psychological aspects of concurrent language processing. The goal is to move towards a more psychologically realistic model of parallel parsing.
Summary updated