Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

The Ecological Fallacy in Annotation: Modelling Human Label Variation goes beyond Sociodemographics

Matthias Orlikowski, Paul Röttger, Philipp Cimiano, Dirk Hovy

TL;DR

This paper investigates whether explicitly modelling annotator sociodemographics via group-specific layers improves prediction of individual annotation decisions for toxic content detection. Using a RoBERTa-based multi-annotator model on the kumar_designing_2021 dataset with four sociodemographic attributes, the authors compare sociodemographic group layers against a baseline and randomized group assignments. They find no statistically significant improvement from including sociodemographic attributes, highlighting the ecological fallacy risk and suggesting that annotation variation arises from factors beyond sociodemographics. The work implies that multi-annotator models can handle many annotators without relying on demographic conditioning, but also calls for exploring interactions (intersectionality) and other attributes to capture the true drivers of annotation decisions.

Abstract

Many NLP tasks exhibit human label variation, where different annotators give different labels to the same texts. This variation is known to depend, at least in part, on the sociodemographics of annotators. Recent research aims to model individual annotator behaviour rather than predicting aggregated labels, and we would expect that sociodemographic information is useful for these models. On the other hand, the ecological fallacy states that aggregate group behaviour, such as the behaviour of the average female annotator, does not necessarily explain individual behaviour. To account for sociodemographics in models of individual annotator behaviour, we introduce group-specific layers to multi-annotator models. In a series of experiments for toxic content detection, we find that explicitly accounting for sociodemographic attributes in this way does not significantly improve model performance. This result shows that individual annotation behaviour depends on much more than just sociodemographics.

The Ecological Fallacy in Annotation: Modelling Human Label Variation goes beyond Sociodemographics

TL;DR

This paper investigates whether explicitly modelling annotator sociodemographics via group-specific layers improves prediction of individual annotation decisions for toxic content detection. Using a RoBERTa-based multi-annotator model on the kumar_designing_2021 dataset with four sociodemographic attributes, the authors compare sociodemographic group layers against a baseline and randomized group assignments. They find no statistically significant improvement from including sociodemographic attributes, highlighting the ecological fallacy risk and suggesting that annotation variation arises from factors beyond sociodemographics. The work implies that multi-annotator models can handle many annotators without relying on demographic conditioning, but also calls for exploring interactions (intersectionality) and other attributes to capture the true drivers of annotation decisions.

Abstract

Many NLP tasks exhibit human label variation, where different annotators give different labels to the same texts. This variation is known to depend, at least in part, on the sociodemographics of annotators. Recent research aims to model individual annotator behaviour rather than predicting aggregated labels, and we would expect that sociodemographic information is useful for these models. On the other hand, the ecological fallacy states that aggregate group behaviour, such as the behaviour of the average female annotator, does not necessarily explain individual behaviour. To account for sociodemographics in models of individual annotator behaviour, we introduce group-specific layers to multi-annotator models. In a series of experiments for toxic content detection, we find that explicitly accounting for sociodemographic attributes in this way does not significantly improve model performance. This result shows that individual annotation behaviour depends on much more than just sociodemographics.
Paper Structure (22 sections, 1 figure, 6 tables)

This paper contains 22 sections, 1 figure, 6 tables.

Table of Contents

  1. Introduction
  2. Contributions
  3. Related Work
  4. Sociodemographics in Annotation Behaviour
  5. Predicting Annotators' Decisions on Text
  6. Data
  7. Experiments
  8. Evaluation Setup
  9. Cross-Validation
  10. Statistical Significance
  11. Remarks on Groups
  12. Results
  13. Sociodemographics vs. Baseline
  14. Sociodemographics vs. Random
  15. Discussion
  16. ...and 7 more sections

Figures (1)

  • Figure 1: Group-specific layers representing annotator sociodemographics in multi-annotator models.