BusTUC - A natural language bus route oracle

Tore AmbleNAACL 2000

TL;DR

This paper introduces BusTUC, a natural language expert system designed to answer public bus route queries. It demonstrates a practical application of natural language processing (NLP) by integrating with a real-world bus company's database, offering a user-friendly alternative to traditional query methods. The system's success highlights the potential for NLP to enhance public services and provide more intuitive access to information.

What Problem Does This Paper Address?

The paper addresses the challenge of making complex information systems, like public transportation databases, accessible to the general public through natural language. Traditionally, users would need to know specific query languages or navigate rigid menu systems. The authors aim to bridge this gap by creating a system that understands and responds to user queries phrased in everyday language, thereby improving user experience and information accessibility. They also aim to prove that a natural language interface can be preferred over interrogative modes for information retrieval, provided it is correct, user-friendly, and comprehensive within its domain.

Methods

BusTUC builds upon the foundation of the CHAT-80 system, a state-of-the-art natural language system developed in the early 1980s. The core methodology involves a pipeline of NLP processes: lexical analysis, syntax analysis, semantic analysis, pragmatic reasoning, and database query processing.

Key methodological aspects include:

  • Adaptability and Reusability: The system is designed with a general grammar and dictionary, with most application-specific changes handled by a tabular semantic knowledge base. This allows for easier adaptation to new domains.
  • Bilingualism: BusTUC supports both Norwegian and English, achieved by maintaining parallel sets of dictionaries, morphology, and grammars. Language detection is based on word frequency.
  • Consensical Grammar: The parser utilizes a variant of Extraposition Grammar called Consensical Grammar. This approach defines incomplete constructs by specifying differences between complete ones, allowing for more intuitive rule definition and handling of movements in sentences.
  • Semantic Type Checking and Heuristics: To manage ambiguity, the parser integrates semantic type checking and employs a heuristic that favors the longest semantically correct phrase.
  • Event Calculus: The system translates natural language queries into a Temporal Query Language (TQL), inspired by Event Calculus, which represents events and their temporal relationships.
  • Production System for Pragmatic Reasoning: A production system named Pragma, with 580 rules, is used to translate TQL expressions into a route database query language (BusLOG), which is essentially a Prolog program.

Datasets

The paper does not explicitly mention specific datasets used for training or evaluation in the traditional machine learning sense. Instead, the system's knowledge base is built from:

  • Bus Route Data: The system is integrated with the actual route database of the Trondheim bus company, containing information on bus lines, stations, and departures.
  • Semantic Knowledge Base: This internal knowledge base contains approximately 4000 entries and defines the semantic relationships and type hierarchies for the domain.
  • Dictionaries and Grammars: The system uses a Norwegian grammar with around 1300 rules and a large name table of 3050 names to handle variations in station naming.

Key Insights and Ideas

  • Practical NLP for Public Services: BusTUC demonstrates that natural language interfaces are not just academic curiosities but can be successfully deployed to enhance public services, making information more accessible and user-friendly.
  • The Power of Domain-Specific Adaptation: By building upon a general NLP framework (TUC) and adapting it with a domain-specific semantic knowledge base and route data, the system achieves high accuracy and relevance within its narrow domain.
  • Bilingualism as a Design Principle: The successful implementation of bilingual support (Norwegian and English) shows that language independence can be achieved through parallel language resources, making the system more broadly applicable.
  • Consensical Grammar's Intuitive Approach: The paper suggests that Consensical Grammar offers a more intuitive way to define grammatical rules, especially for handling sentence movements, compared to traditional grammar formalisms.
  • The Importance of Pragmatic Reasoning: The explicit use of a production system (Pragma) for pragmatic reasoning highlights its crucial role in translating semantic representations into executable database queries.

Key Results

  • Operational System: BusTUC has been operational on the Internet since 1999 and was adopted by the local bus company in Trondheim.
  • High Volume of Queries: The system has answered over 150,000 questions, indicating its popularity and utility.
  • Low Error Rate: The error rate for single, correct, complete, and relevant questions is reported to be about 2 percent.
  • Fast Response Times: Average response time is usually less than 2 seconds, with some complex queries taking up to 10 seconds.
  • Successful Domain Transfer: The underlying TUC framework, initially developed for geography and hospital administration, proved adaptable to the bus route domain.

Limitations and Future Work

  • Parsing Efficiency and Language Coverage: The authors acknowledge that further work is needed to improve parser efficiency and expand language coverage to handle a wider range of expressions and ambiguities.
  • Timeout on Parsing: The parsing process has a timeout of 10 seconds, indicating that some legal sentences can still be computationally intensive to parse, even with optimizations.
  • Ambiguity Resolution: While heuristics and semantic type checking are used, disambiguation remains a significant challenge, and some committed choices in the grammar can lead to unrecoverable errors.
  • Portability and Scalability: The paper raises the question of whether such systems will become a preferred way of offering information and notes that significant work would be required to make the system portable and implementable elsewhere, potentially connecting to various travel agencies.
  • Future Interest: There is growing interest from other bus companies and route information providers for similar systems, suggesting a potential for wider adoption.

Summary updated

BibTeX
@inproceedings{amble2000bustuc,
  author = {Tore Amble},
  title = {{BusTUC - A natural language bus route oracle}},
  booktitle = {Conference of the North American Chapter of the Association for Computational Linguistics},
  year = {2000},
}
Figure