Crowdsourced comparative judgement for evaluating learner texts: How reliable are judges recruited from an online crowdsourcing platform?
Abstract
Recent studies of proficiency measurement and reporting practices in applied linguists have revealed widespread use of unsatisfactory practices such as the use of proxy measures of proficiency in place of explicit tests. Learner corpus research is one specific area affected by this problem: few learner corpora contain reliable, valid evaluations of text proficiency. This has led to calls for the development of new L2 writing proficiency measures for use in research contexts. Answering this call, a recent study by Paquot et al. (2022) generated assessments of learner corpus texts using a community-driven approach in which judges, recruited from the linguistic community, conducted assessments using comparative judgement. Although the approach generated reliable assessments, its practical use is limited because linguists are not always available to contribute to data collections. This paper, therefore, explores an alternative approach, in which judges are recruited through a crowdsourcing platform. We find that assessments generated in this way can reach near identical levels of reliability and concurrent validity to those produced by members of the linguistic community.
Study specs
Judges recruited via an online crowdsourcing platform conducted comparative judgement assessments of learner texts to measure writing proficiency.
- Authors
- P Thwaites,N Vandeweerd,M Paquot
- Institution
- University College Londonouvain,Radboud University Nijmegen,Fonds de la Recherche Scientifique – FNRS
- Discipline
- Applied Linguistics
- Study Type
- Evaluation Study
- Year
- 2025
- Human Data Platform
- Prolific
- Source
- View Source Google Scholar
Measured Outcomes
Reliability and concurrent validity of learner text evaluations performed via crowdsourced judges compared to linguist evaluations.
Peer Review & Critical Discussion
Potential Selection Bias in 2023 Cohort
The participant pool shows a concerning overrepresentation of users from high-income demographics. Looking at Table 3, we can see that 78% of respondents had annual incomes above $75k, which significantly limits the generalizability of these findings to broader populations.
Non-naive Participants Issue
I've noticed a methodological concern regarding participant naivety. Given that Prolific users often complete multiple studies, there's a real risk that participants had prior exposure to similar experimental paradigms, which could confound the results.
RLHF Applicability to This Study Design
The implications for RLHF training pipelines are understated. If we accept the authors' conclusions about preference stability, this has direct consequences for how we should structure reward model training. The temporal decay effect described in Section 4.2 is particularly relevant.
Verify your expertise to join discussion
Create an account and verify your credentials to participate in peer discussions.