Panel on Natural Language and Databases

Daniel P. FlickingerACL 1984

TL;DR

This paper argues that natural language interfaces (NLIs) for databases are still a valuable research area, but proposes extending them to interact with programming environments. This shift offers richer linguistic challenges and practical benefits, moving beyond simple data retrieval to system manipulation and customization.

What Problem Does This Paper Address?

The paper addresses the ongoing debate about the utility of natural language interfaces (NLIs) for database querying. While some might consider this a solved problem or a less exciting research avenue, the author contends that significant linguistic challenges remain. Furthermore, the paper identifies a limitation in current NLI research by focusing solely on data retrieval, proposing that a more compelling and practical direction lies in enabling natural language interaction with complex programming environments. This would move beyond simply asking questions about data to actively controlling and modifying software systems.

Methods

The paper doesn't detail specific algorithms or novel techniques in the traditional sense. Instead, it proposes a conceptual shift in the application of NLIs. The core "method" is the idea of extending NLIs from database query systems to programming environments. This involves:

  • Leveraging existing database query strengths: The author acknowledges the merits of database query as a testbed, particularly for issues like language independence and handling ambiguity, where speed is less critical.
  • Expanding to programming environments: The proposed extension involves allowing natural language to not only retrieve information about a system's state but also to modify it. This implies a need for more sophisticated language understanding and potentially language generation capabilities.
  • Focus on linguistic challenges: The paper highlights the linguistic issues that arise from this extension, such as handling new syntactic constructions, managing discourse representation, and dealing with tense and aspect in a more dynamic context.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Database query NLIs are far from solved: Despite advancements, there are still significant linguistic problems to tackle in making NLIs for databases robust, especially concerning language independence and handling ambiguity.
  • Programming environments are a richer testbed: Interacting with a programming environment offers a more challenging and practical domain for NLI research than simple database queries.
  • Beyond retrieval to manipulation: The true power of NLIs lies in enabling users to control and modify systems, not just query them.
  • Language independence is a key metric: The degree to which an NLI can be generalized across languages is a valuable measure of its sophistication.
  • Self-improvement potential: A natural language interface developed within a programming environment could potentially be used to develop the NLI itself, creating an elegant feedback loop.

Key Results

The paper is primarily a position statement and a proposal for future research directions, rather than a report on experimental results. Therefore, there are no quantitative results to summarize. The "key results" are conceptual:

  • The argument that database query NLIs still present linguistically interesting problems.
  • The proposal that extending NLIs to programming environments is a fruitful and necessary next step for the field.
  • The identification of specific linguistic challenges (discourse representation, tense/aspect, new syntactic constructions) that would arise from this extension.

Limitations and Future Work

Limitations:

  • Speed concerns: While the author initially downplays speed in database queries, they acknowledge that for interactive programming environments, processing speed will become a more significant concern.
  • Implementation details: The paper does not delve into the technical specifics of how to merge database-like information storage with general programming environments.
  • User adaptation: The author notes that for common tasks, a natural language interface might be slower than direct manipulation for experienced users.

Future Work:

  • Developing NLIs for programming environments: This is the central call to action.
  • Research into language production: The proposed extended interface would necessitate more sophisticated language production capabilities.
  • Discourse representation research: The dynamic nature of programming environments would provide a natural setting for studying discourse.
  • Tense and aspect research: These linguistic phenomena are more readily explored in the context of system changes over time.
  • Investigating language universals: Further research into how to build language-independent NLIs.
  • Exploring the "how-to" question: Using NLIs to access and present on-line documentation.

Summary updated

BibTeX
@inproceedings{flickinger1984panel,
  author = {Daniel P. Flickinger},
  title = {{Panel on Natural Language and Databases}},
  booktitle = {Annual Meeting of the Association for Computational Linguistics},
  year = {1984},
}
Figure