The Register-specific Distribution of Personification in Hungarian: A Corpus-driven Analysis

Gabor SimonNAACL 2024

TL;DR

This paper introduces PerSE, an extended corpus for analyzing personification in Hungarian across different text types. By applying a corpus-driven approach, the study reveals significant differences in how personification is used in journalistic (car reviews, interstate relations) versus academic texts, offering new insights into the linguistic patterns of this common figurative device.

What Problem Does This Paper Address?

Personification, the attribution of human characteristics to non-human entities, is a pervasive feature of language. While various conceptual models exist to explain it, a comprehensive, corpus-driven analysis of its linguistic realization, particularly its variability across different text types (registers), has been lacking. Existing research often focuses on specific aspects or uses limited datasets. This paper aims to bridge this gap by providing a detailed analysis of personification in Hungarian, exploring how its usage, structure, and semantic quality differ across journalistic and academic registers.

Methods

The study employs a corpus-driven methodology, building upon an extended version of the PerSE corpus. The core of the method involves a semi-automatic and manual annotation process. This includes:

  • Corpus Building: Compiling texts from three distinct registers: car reviews, reports on interstate relations, and academic research papers.
  • Annotation Protocol: Utilizing a protocol inspired by existing metaphor identification methods (like MIPVU and MetaID), which labels personification components (node words and arguments), semantic relations, and categorizes personifications into four types: conventional, novel, default, and metonymic.
  • Lexicogrammatical Analysis: Examining the frequency and distribution of personifications, as well as analyzing the personifying and non-personifying use of frequent verbs across registers.
  • Statistical Analysis: Employing non-parametric statistical tests (Wilcoxon tests, ANOVA, Chi-squared tests) to identify significant differences in personification patterns between registers.

Datasets

The primary dataset is the PerSE corpus, an extended version specifically designed for studying personification in Hungarian. It comprises:

  • Car reviews: 6 texts, 10,468 tokens.
  • Reports on interstate relations: 5 texts, 7,938 tokens.
  • Academic research papers: 5 texts, 11,500 tokens.

This corpus is a new resource, extending previous work by including multiple registers and a more comprehensive annotation scheme.

Key Insights and Ideas

  • Register-Specific Personification: The study strongly suggests that personification is not used uniformly across different text types. Journalistic registers, particularly reports on interstate relations, tend to use personification more frequently and in more elaborate ways than academic texts.
  • Complexity and Idiomaticity: Interstate relations reports show a higher proportion of idiomatic personifications and more complex argument structures (more arguments per node) compared to car reviews. This indicates a preference for established, prefabricated expressions in political discourse.
  • Novelty in Car Reviews: Car reviews, on the other hand, exhibit a higher frequency of novel personifications, suggesting a creative use of the device to describe technical objects.
  • Verb-Specific Register Preferences: Certain common verbs show distinct patterns of personifying use depending on the register. For example, the verb "tud" (know/can) is more frequently personified in car reviews, likely due to describing the "capacities" of vehicles.
  • Academic Texts' Role: While academic texts show lower overall personification frequency, they dominate in conventional personifications, suggesting a more established and less creative use of the phenomenon.

Key Results

  • Frequency: Reports on interstate relations had the highest frequency of personifications (34,759.33 pmw), followed by car reviews (33,985.32 pmw), and then research papers (28,267.87 pmw). This is the reverse of some previous metaphor studies.
  • Idiomaticity: Interstate reports had a significantly higher number of idiomatic personifications than car reviews (p<0.05).
  • Novelty: Car reviews used significantly more novel personifications than the other two registers (p<0.001).
  • Metonymic Personifications: Interstate reports contained significantly more metonymic personifications than car reviews (p<0.001).
  • Verb "tud": A significant association was found between the personifying use of the verb "tud" and the register (x²(2)=14.841, p<0.001), with car reviews showing a higher preference for its personified use.

Limitations and Future Work

  • Corpus Size and Availability: The PerSE corpus, while extended, is still considered a modest-scale research corpus. The authors emphasize the need to make it publicly available and to continue sampling more texts, particularly literary fiction and conversations.
  • Annotation Refinement: Further refinement of the annotation protocol is necessary to increase inter-annotator agreement, especially for classifying the semantic quality of personifications.
  • Validation: The current dictionary-based analysis demonstrates feasibility, but the actual comprehension of these personifications by readers needs to be tested through alternative empirical methods (experimental, questionnaire-based).
  • NLP Applications: The corpus holds potential for training NLP models for automatic personification detection, but the effectiveness of such models remains to be seen.
  • Future Research: The authors suggest expanding the corpus to include literary texts, exploring the construction-like behavior of personification, and using the corpus as a benchmark for cross-linguistic studies. They also propose that personification should be viewed as a complex phenomenon in its own right, not just a subtype of metaphor.

Summary updated

BibTeX
@inproceedings{simon2024register,
  author = {Gabor Simon},
  title = {{The Register-specific Distribution of Personification in Hungarian: A Corpus-driven Analysis}},
  booktitle = {Conference of the North American Chapter of the Association for Computational Linguistics},
  year = {2024},
}
Figure