Answering Why and How questions with respect to a frame-based knowledge base: a preliminary report

Chitta Baral, Nguyen Ha Vo, Shanshan Liang

TL;DR

This paper tackles the challenge of answering "why" and "how" questions, which are often overlooked in favor of factoid questions in current QA systems. The authors propose a novel approach using frame-based knowledge bases and Answer Set Programming (ASP) to formally define and answer these complex question types, specifically focusing on biological processes.

What Problem Does This Paper Address?

Traditional question answering (QA) systems have made significant strides in retrieving factual answers (factoid questions). However, answering "why" and "how" questions, which require understanding causality, processes, and importance, remains a significant challenge. These types of questions are fundamental to human understanding, even at the primary school level, and are crucial for applications like intelligent tutoring systems. The paper identifies a gap in research concerning how to formally represent and reason with knowledge bases to answer these more nuanced questions, particularly within structured knowledge representations like frame-based systems.

Methods

The core of the paper's methodology lies in:

  • Frame-Based Knowledge Representation: The authors leverage a frame-based knowledge base (KB), specifically the AURA biology KB, which represents classes, objects, and their relationships through slots.
  • Event Description Graphs (EDGs): To provide a more abstract and visual representation of biological processes, they introduce EDGs. These directed graphs represent events and entities, with compositional and behavioral edges illustrating their relationships.
  • Formalizing Answers: They formally define what constitutes an answer to two specific question types:
  • "How are X and Y related in the process Z?": The answer is represented as a minimal subgraph (MIN_EDG) of the EDG that captures the essential connections between X and Y within process Z, focusing on the lowest common ancestor and relevant paths.
  • "Why is X important to Y?": The answer combines the MIN_EDG with an "important path" (ipath) to highlight the causal or functional significance of X to Y.
  • Answer Set Programming (ASP): The formal definitions and reasoning processes are implemented using ASP. ASP is chosen for its strong theoretical foundation, expressiveness, and suitability for declarative knowledge representation and reasoning, particularly for handling inheritance and complex logical rules.

Datasets

The paper uses the AURA knowledge base, a manually curated frame-based KB containing extensive information on biology concepts and processes. While not a benchmark dataset in the traditional sense, AURA serves as the primary source of knowledge for their experiments and examples. The authors mention that examples are sometimes modified slightly for clarity.

Key Insights and Ideas

  • Formalizing "Why" and "How": The paper's most significant contribution is the formal definition of answers to "why" and "how" questions within a structured knowledge base. This moves beyond simple keyword matching or fact extraction.
  • Event Description Graphs as a Core Abstraction: EDGs provide a powerful visual and structural representation that bridges the gap between the detailed frame-based KB and the abstract reasoning required for these questions.
  • ASP for Declarative Reasoning: The successful application of ASP demonstrates its utility for implementing complex reasoning tasks like inheritance, cloning, and pathfinding, which are essential for answering sophisticated questions.
  • Focus on Biological Processes: By grounding their work in the biology domain, the authors highlight the practical applicability of their approach to complex scientific reasoning.

Key Results

The paper presents a formal framework and an ASP implementation for answering two specific types of "why" and "how" questions. While quantitative results are not the primary focus of this preliminary report, the paper demonstrates through examples (photosynthesis, event description graphs) how their approach can generate meaningful answers by constructing specific subgraphs (MIN_EDG) and identifying relevant paths. The ASP encoding provides a declarative way to derive these answers.

Limitations and Future Work

  • Preliminary Report: The authors explicitly state this is a preliminary report, suggesting that the work is ongoing.
  • Limited Question Types: The current framework is focused on two specific question types: "How are X and Y related in the process Z?" and "Why is X important to Y?". The paper acknowledges that many other "why" and "how" question forms exist.
  • Future Work: The authors plan to extend their approach to handle additional types of "why" and "how" questions. They provide examples of how their framework could be adapted for questions like "How does X occur?" or "How does X produce Y?". They also mention the need for further propositions to formally relate their abstract definitions to the ASP-derived answers.

Summary updated

Abstract

Being able to answer questions with respect to a given text is the cornerstone of language understanding and at the primary school level students are taught how to answer various kinds of questions including why and how questions. In the building of automated question answering systems the focus so far has been more on factoid questions and comparatively little attention has been devoted to answering why and how questions. In this paper we explore answering why and how questions with respect to a frame-based knowledge base and give algorithms and ASP (answer set programming) implementation to answer two classes of questions in the Biology domain. They are of the form: "How are X and Y related in the process Z?" and "Why is X important to Y?"
BibTeX
@inproceedings{baral2012answering,
  author = {Chitta Baral and Nguyen Ha Vo and Shanshan Liang},
  title = {{Answering Why and How questions with respect to a frame-based knowledge base: a preliminary report}},
  booktitle = {ICLP (Technical Communications)},
  year = {2012},
  doi = {10.4230/lipics.iclp.2012.26},
}

Loading related content...

Figure