Analysis of Titles and Readers for Title Generation Centered on the Readers

Yasuko Senda, Yaushi ShinoharaCOLING 2002

TL;DR

This paper argues that traditional title generation, focused on summarizing content, often fails to engage readers. By analyzing how newspaper headlines differ from technical paper titles, they identify reader-centric strategies. Their survey reveals that titles emphasizing the purpose of a technology are more effective for general audiences, suggesting a shift from author-centric to reader-centric title generation.

What Problem Does This Paper Address?

Generating effective titles for technical documents is a challenge. While existing methods focus on extracting keywords to create a concise summary (an "author-centered approach"), they often neglect a crucial aspect: attracting the reader's interest. A title that accurately summarizes but fails to engage can lead to disappointment or simply be ignored. This paper addresses the gap by investigating how to create titles that not only inform but also entice a general audience to read the document, proposing a "reader-centered approach."

Methods

The authors employ a two-pronged approach:

  1. Comparative Analysis of Titles and Headlines: They analyzed Japanese technical paper titles and corresponding newspaper headlines. To do this, they developed a set of "syntactic function tags" (e.g., Behavior, Object, Purpose, Method, Strong Point) to categorize text segments within titles and headlines. This allowed them to compare the typical content and wording used in each.
  2. Questionnaire Survey: To empirically validate their findings, they conducted a survey. They generated titles for newly developed technologies using different expression patterns identified in the comparative analysis. Respondents, categorized by their interest and expertise level in specific technical fields, were asked about the comprehensibility, positive feelings evoked, and interest generated by these titles.

Datasets

The study utilized two main sources of data:

  • Technical Papers and Newspaper Articles: Titles and headlines were collected from a research institute's database, covering science and technology. Approximately 90% of newspaper headlines had a corresponding technical paper.
  • Generated Titles for Survey: The authors created new titles based on identified expression patterns for three technical fields (electric transmission, architectural engineering, and environmental science). These were used in the questionnaire.

Key Insights and Ideas

  • The "Author-Centered" vs. "Reader-Centered" Divide: Traditional title generation focuses on what the author wants to say (keywords, summary). This paper advocates for understanding what the reader needs to see to be engaged.
  • Newspaper Headlines as a Model for Reader Engagement: Headlines, unlike technical titles, are designed to grab attention. Their analysis revealed that headlines often prioritize the purpose of a technology or its advantages over technical details.
  • Purpose Over Process for General Audiences: For general readers, titles that clearly state the purpose of a technology (e.g., "for power distribution cables") are significantly more effective at generating comprehension, positive feelings, and interest than those detailing the method or technical jargon.
  • Expertise Matters: While purpose-driven titles are generally effective, the impact diminishes for readers with higher technical expertise. For these readers, titles that describe the technology's content (what it does) might also be important.
  • Obligatory vs. Optional Components: The study identified "obligatory" components (Behavior, Object, Technology Type) that are essential for conveying substance, and "optional" components (Purpose, Method, Strong Point) that can be used to tailor the title for reader engagement.

Key Results

  • Purpose-driven titles (Pattern 2.0) were most effective: Across most readership groups, titles emphasizing the purpose of the technology were rated highest for comprehensibility, evoking positive feelings, and generating interest.
  • Expertise dampens the effect: The strength of the association between title expression patterns and reader impressions was weaker for "engineer" and "researcher" groups compared to "unconcerned" or "commoner" groups.
  • Chi-square tests confirmed significant associations: The study found statistically significant relationships between title expression patterns and reader impressions for all readership categories.

Limitations and Future Work

  • Limitations: The authors acknowledge that the strength of associations between title patterns and impressions is lower for readers with higher expertise. This suggests that a single "best" pattern might not exist for all audiences. The study also focused on specific technical fields, and findings might vary across other domains.
  • Future Work: The authors plan to further investigate the impact of "optional components" on reader impressions. They also aim to develop a method for automatically generating titles by extracting and combining phrases from the body text, aligning with the identified effective expression patterns.

Summary updated

BibTeX
@inproceedings{senda2002analysis,
  author = {Yasuko Senda and Yaushi Shinohara},
  title = {{Analysis of Titles and Readers for Title Generation Centered on the Readers}},
  booktitle = {International Conference on Computational Linguistics},
  year = {2002},
}
Figure