Zone analysis in biology articles as a basis for information extraction

Summary

In the field of biomedicine, an overwhelming amount of experimental data has become available as a result of the high throughput of research in this domain. The amount of results reported has now grown beyond the limits of what can be managed by manual means. This makes it increasingly difficult for the researchers in this area to keep up with the latest developments. Information extraction (IE) in the biological domain aims to provide an effective automatic means to dynamically manage the information contained in archived journal articles and abstract collections and thus help researchers in their work. However, while considerable advances have been made in certain areas of IE, pinpointing and organizing factual information (such as experimental results) remains a challenge. In this paper we propose tackling this task by incorporating into IE information about rhetorical zones, i.e. classification of spans of text in terms of argumentation and intellectual attribution. As the first step towards this goal, we introduce a scheme for annotating biological texts for rhetorical zones and provide a qualitative and quantitative analysis of the data annotated according to this scheme. We also discuss our preliminary research on automatic zone analysis, and its incorporation into our IE framework.

Introduction

Information extraction (IE) in the biological domain is now regarded as an essential technique for utilizing information contained in archived journal articles and abstract collections such as MEDLINE. Major domain databases offer large-scale archives of semi-structured results, but they tend to be incomplete or not up-to-date. For the latest results, as well as for confirmation of results reported in the database and for supplementary information, access to the latest literature is necessary. Thus, the significance of a more sensible management of facts, specifically an integration and update of experimental results, is self-evident. Given the limitations of manual work for such purposes in terms of both efficiency and accuracy, IE's focus on factual information is of a critical importance.

Recent intensive research in natural language processing in the biological domain (bioNLP) has made major progress in the extraction of bio-named entities and biological interactions (e.g. [1], [2], [3]), but further advancement aimed at pinpointing and organizing factual information remains a challenge. In particular, the important task of identifying new experimental results is complicated by the large number of statements made in each article that pertain to results in general, including references to previous work as well as technical details and conjectures. The same information (about a molecular event, for example) may be provided as a new result, as a previously known result, as a conjecture, etc., that is, in different rhetorical contexts. Because current IE relies on surface lexical and syntactic patterns, it is not sensitive to the rhetorical status of information. As a consequence, existing techniques tend to extract old results mixed with new ones, leaving the novel contribution unclear. We expect that a rhetorical analysis of biological texts and its incorporation into IE will provide a means to address this problem.

As the first step towards this goal, we proposed in [4] annotating biology texts in terms of rhetorical zones with a shallow nesting using a scheme modified from [5]. In [6], we gave a qualitative analysis of our sample hand-annotated data and described how such zones can be identified by humans. In this paper, we provide a comprehensive qualitative and quantitative analysis of the process and the results of zone analysis (ZA) in the hand-annotated data, describe some preliminary work on automatic ZA and discuss its future incorporation into our framework for IE.

The organization of the paper is as follows. We first discuss in more detail the motivations for ZA in biology, from an IE perspective and in the view of previous work (Section 2). We then introduce our framework. We describe general characteristics of biological articles and introduce our annotation scheme and annotated data (Section 3). We then shed light on the decision process in which zones can be identified by a human annotator and describe, mostly qualitatively, the main features required for identification of each zone class (Section 4). To summarize the results of the annotation work, we illustrate the distribution of zones in the annotated articles both visually and quantitatively (Section 5). Finally, we discuss ongoing and future work on automatic ZA and its integration to IE (Section 6).