Character-level embeddings increase robustness to student misspellings, but spelling correction as pre-processing was found more effective¶

Claim · All claims
Evidence · 1 study (2 entries) · review r2 · q2
Currency · newest study 2023 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Models incorporating character embeddings have been shown to be more robust against spelling errors in students' texts (Horbach et al., 2017). → Xiaoyu Bai · Manfred Stede 2023 q2 i? Riordan et al. (2019) found character-level information showed positive effects but was not as effective as performing spelling correction on the training data as a pre-processing step. → Xiaoyu Bai · Manfred Stede 2023 (2)

Evidence¶

Xiaoyu Bai · Manfred Stede 2023¶

Xiaoyu Bai · Manfred Stede. (2023). A Survey of Current Machine Learning Approaches to Student Free-Text Evaluation for Intelligent Tutoring. International Journal of Artificial Intelligence in Education (2023) 33:994–1032. https://doi.org/10.1007/s40593-022-00323-0

q2 · i? · review · r2

The survey reports, citing Horbach et al. (2017), that character embeddings capture the relatedness between a word and its misspelled counterpart because they share many substrings. No effect size is printed.

"models incorporating character embeddings have been shown to be more robust against spelling errors in students' texts (Horbach et al., 2017)"

Xiaoyu Bai · Manfred Stede 2023 (2)¶

Xiaoyu Bai · Manfred Stede. (2023). A Survey of Current Machine Learning Approaches to Student Free-Text Evaluation for Intelligent Tutoring. International Journal of Artificial Intelligence in Education (2023) 33:994–1032. https://doi.org/10.1007/s40593-022-00323-0

q2 · i? · review · r2

The survey reports Riordan et al. (2019)'s GRU-based short-answer scoring studies with character-level information; the survey states the benefits of character-level embeddings for misspellings are inconclusive. No effect size is printed.

"Riordan et al. ( 2019) found in their studies that while they did show positive effects, they were not as effective as performing spelling correction on the training data as a pre-processing step."

Discussion¶

Related Claims¶

• Spellcheck and autocorrect are not foolproof: efficacy is bounded near 80% and the most common student writing error is now correctly spelled but inappropriate words — related
• Holistic scoring of German university students' essays posed significant challenges to both a neural model and an SVM with n-gram features — related