The Scientization of Literary Study

Stefania Degaetano-Ortlieb, Andrew PiperNAACL 2019

TL;DR

This paper investigates whether literary studies, traditionally seen as distinct from science, has undergone a "scientization" process over the past half-century. Using information-theoretic measures on large corpora of academic texts, they find evidence that literary studies has indeed become more linguistically similar to scientific disciplines, though it still retains a closer connection to standard English.

What Problem Does This Paper Address?

The humanities, and literary studies in particular, have long been characterized by their qualitative methods and distinct epistemological stance, setting them apart from the natural and social sciences. This perception is often reinforced by the idea that scientific disciplines are inherently more rigorous and objective. However, this paper challenges this dichotomy by proposing that literary studies, like other academic fields, might be undergoing a process of "scientization." This isn't about adopting scientific jargon, but rather about linguistic shifts that align it more with scientific discourse. The authors aim to quantify this shift by examining three specific linguistic practices: social differentiation (how a field's language diverges from common language), diachronic specialization (how a field's language evolves and becomes more expert-oriented over time), and phrasal standardization (the increased predictability of word sequences).

Methods

The core of the methodology relies on information-theoretic measures to analyze linguistic divergence and standardization.

  • Kullback-Leibler Divergence (KLD): This measure quantifies how much one probability distribution differs from another. In this context, it's used to compare the language models of different corpora (literary studies vs. standard English vs. scientific English) and to track changes within literary studies over time. A higher KLD value indicates greater divergence.
  • Hypothesis 1 (Social Differentiation): They test if literary studies diverges less from standard English than scientific language does.
  • Hypothesis 2 (Diachronic Specialization): They examine if literary studies' language becomes more distinct from its past self over time, and if this divergence is asymmetrical (i.e., the present is more distinct from the past than vice versa).
  • Surprisal: This measure quantifies the information content of a linguistic unit (like a word or phrase) in a given context. Low surprisal indicates high predictability and thus a more standardized linguistic unit.
  • Hypothesis 3 (Phrasal Standardization): They use surprisal on trigrams (sequences of three words) to measure the degree of standardized phrases in literary studies over time. A decrease in average surprisal per document suggests an increase in standardized phrases.

Datasets

  • Literary Research Article Corpus (LRA): This is the primary dataset for literary studies, comprising approximately 63,400 articles published between 1950 and 2010 from 60 academic journals.
  • Corpus of Historical American English (COHA): Used to represent standard American English. It's a large corpus (over 400 million words) spanning from the 1810s to the 2000s, balanced across genres. For this study, texts from 1950 onwards were used to align with the LRA corpus.
  • Royal Society Corpus (RSC): Used to represent scientific English. This corpus contains journal publications from the Royal Society of London, with texts from 1665 to 1996. For this study, texts from 1950 to 1996 were used to match the LRA corpus.

Key Insights and Ideas

  • Literary Studies is Not Static: The traditional view of literary studies as a fixed, qualitative discipline is challenged by evidence of linguistic evolution.
  • "Scientization" is Nuanced: The process isn't a simple adoption of scientific jargon but a subtler shift in linguistic patterns, affecting differentiation, specialization, and standardization.
  • A "Middle Ground" Discipline: Literary studies appears to occupy an interesting position, diverging from standard English but not as much as scientific language does, suggesting a unique role as a mediator between academic discourse and the public.
  • Asymmetrical Change: The finding that recent literary texts are better modeled by older texts than vice versa (and vice versa for science) points to a genuine evolution of specialized vocabulary and concepts within the field.

Key Results

  • Social Differentiation (H1): Literary studies articles diverge less from standard English than scientific articles do, supporting the idea that literary studies remains closer to common language.
  • Diachronic Specialization (H2):
    • From the 1950s to the mid-1970s, literary studies was more similar to standard English than to scientific language.
  • After the mid-1970s, literary studies began to diverge more equally from both standard and scientific English, showing a trend towards specialization.
  • The divergence between past and present literary texts was asymmetrical, with recent texts being less predictable from older texts, indicating the development of field-specific language. This asymmetry was more pronounced in literary studies than in standard English.
  • Phrasal Standardization (H3):
    • Literary studies articles show a significant decrease in average surprisal over time, indicating an increase in standardized phrases.
  • While this trend is present, the percentage of standardized phrases in literary studies is lower than in scientific articles, reinforcing its position as less "scientific" than hard sciences.

Limitations and Future Work

  • Geographic Scope: The study is limited to an Anglo-Saxon context and does not account for disciplinary practices in other national frameworks.
  • Text Type: The analysis focuses solely on articles, excluding monographs, which are also a significant genre in literary studies.
  • Mechanisms of Scientization: The paper identifies the process of scientization but does not explain the drivers behind it (e.g., editorial policies, institutional pressures, demographic shifts). Future research is needed to explore these mechanisms.
  • Broader Linguistic Features: The study could be expanded to include other linguistic features like citation practices, evidentiary norms, or the use of word embeddings for richer contextual analysis.
  • Vocabulary Expansion: Further exploration of how vocabulary and punctuation changes contribute to grammatical consolidation is suggested.

Summary updated

BibTeX
@inproceedings{degaetanoortlieb2019scientization,
  author = {Stefania Degaetano-Ortlieb and Andrew Piper},
  title = {{The Scientization of Literary Study}},
  booktitle = {Conference of the North American Chapter of the Association for Computational Linguistics},
  year = {2019},
}
Figure