Automated Essay Scoring Using Transformer Models
Problem
Manual scoring of learner text is burdensome; the study tests whether transformer models beat bag-of-words methods for automated scoring.
Why it matters
Automated scoring reduces manual scoring burden and enables ad hoc feedback for learners, per the abstract.
What the AI did
Transformer models classified politeness in 2088 manually labeled email responses and were compared with a bag-of-words logistic regression baseline.
Results
Both transformer models outperformed the regression-based baseline without hyperparameter tuning; no numeric metrics stated.
Impact
On this single dataset, transformers showed significant advantages over bag-of-words for politeness-classification AES tasks.
Record
| Venue | Psych |
|---|---|
| Published | 2021 |
| Work type | article |
| Authors | Sabrina Ludwig, Christian Mayer, Christopher Hansen, Kerstin Eilers, Steffen Brandt |
| Citations | 9 (OpenAlex, at import time) |
| References | 93 |
| Model or method | Transformer models (specific variants not stated in abstract) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.3390/psych3040056 |
| Open access | gold |
| Keywords | automated essay scoring, transformers, text classification, politeness, natural language processing |
Importance rating
real but narrower advance
Abstract states automated scoring significantly reduces manual scoring burden and enables ad hoc learner feedback; benefit is real but confined to education scoring workflows; no scale given.