Faculdade de Letras da Universidade do Porto - OCS, 15th INTERNATIONAL ISKO CONFERENCE

Font Size: 
Wei-Ning Cheng, Christopher S. G. Khoo

Last modified: 2018-02-15


Research papers typically have an abstract that is structured in a variation of the well-known IMRaD structure: introduction, method, results and discussion/conclusion/contribution. Content-wise, an abstract gives an overview of the research topic, but often also specifies the research objectives, and outlines the methods and research results. A good abstract will also make an argument (often indirectly) for why the research objective is worth investigating (e.g., by pointing out a research gap), and for the validity of the research results and significance/contribution of the study. The content (i.e. information) and arguments have to be expressed in text in a way that helps the reader to understand the information provided and to be persuaded of the arguments.

This study analyzes the information structure that support the following types of arguments found in sociology research abstracts: research objective, research result, research gap, and significance/contribution of the research. An argument can be divided into the argument point or claim (e.g., a research objective) and supporting argument (e.g., information that justifies the research objective). The argument point can be analyzed into component concept-relation-concept triples, that indicate the various types of information that make up the argument point. The supporting argument can also be analyzed into concept-relation-concept triples. The concept-relation-concept triples are linked up into an information structure or pattern.

The data for this study were abstracts taken from 24 journal articles published in eight sociology journals with high impact factors listed in InCites Journal Citation Reports—three articles from each journal. The abstracts were analyzed to identify common concept-relation-concept triples and bigger information patterns in the text content, especially in the research objective, research result, research gap and significance/contribution statements.

We developed a few semantic frames to represent common information structures or patterns, following frame semantics theory. This paper describes three common semantic frames found in the research abstracts:

  • the Research-relation frame represents the structure of research objectives or research results that identify an association (e.g., causal relation) between two concepts, as well as the modality (i.e. true, false, probable, etc.) and size of the relation. It also represents other types of information related to the argument (i.e. context, qualifier and rebuttal) as well as the supporting argument (i.e. evidence and explanation).
  • the Comparison frame represents comparisons between two or more concepts as well as the result of the comparison.
  • the Study frame represents the research structure at a high level—the type of research objective, type of research method and type of research result. In particular, it indicates whether the study represents a historical, descriptive, development, evaluative or investigative type of study. For example, a development type of study seeks to develop a method, theory/model or software system.

The three semantic frames with their associated semantic roles (indicating different types of related information) are shown graphically in Figures 1 to 3.

The results of applying these frames to analyze the information structure in the four types of arguments will be reported, including the frequency distributions of the three semantic frames and their semantic roles. For example, the most commonly used Research-relations in the 24 sociology research abstracts are Cause-effect and Association, each occurring in 67% of the abstracts. The abstracts typically highlight the research context (50%) and specify a limitation (29%) related to the research context (e.g., the location of data collection).

As for the types of study, the 24 sociology research abstracts include 3 historical analyses, 5 development, 8 evaluation, 6 identification, and 14 investigation studies. Some studies involve more than one type of research.

The study also analyzed the relations among the three semantic frames. Thus, the information structures in research abstracts can be represented as a network of instantiated semantic frames. For example, a Research-relation instance is often linked to a Comparison instance, as research study often involves comparison between two or more things. A study may compare two subclasses of a concept defined by differences in an attribute, for example hotel websites with and without embedded social media channels. The comparison may focus on (or measure) another attribute of the concept, for example comparing user experience for hotel websites with and without embedded social media channels.

A comparison may also be made between the study results and a hypothesis, an expectation, or what has been reported in previous studies. For example, in this statement “This indicates that the ability of messages to cross small-world human social networks may be overestimated”, implies that the result is less than what was expected. In addition, investigation of a Research_relations sometimes involve comparisons of attributes of the Research _relation.

The relevance or relation of the research to the broader field and to previous research is sometimes made, and this often involves an implicit or explicit comparison, as in the following example, “This study extended this line of research to investigate whether these same maternal education patterns in parenting are observed among a set of parenting behaviors that are linked to young children’s health.”

Filling a research gap is one of the most important and prevalent argument. A research gap can be expressed in terms of the Research_relation. An author can highlight the gap by indicating that the relation between two concepts is not fully known, and it is desirable to find out. For example: “Little is known about couples’ shared time and how actual time spent together is associated with well-being.” The Comparison frame can also be used to characterize a research gap by comparing two subclasses of the concepts involved, two attributes of a concept, or two or more values of a Research_relation attribute.

The motivation for this study is to identify common information structures and arguments for the purpose of teaching students how to write good research abstracts. However, the semantic frames and their relations are expected to be useful for developing knowledge representation schemes to support text summarization of research papers.

Keywords: Semantic frames, Argument structures, Knowledge organization