CoChat

A Semantic Role-based Approach to Open-Domain Automatic Question Generation

Open access
Flor, Michael, Riordan, Brian · 2018 · Proceedings of the Thirteenth Workshop on Innovative Use of NLP for Building Educational Applications

The DOI matches its Crossref record; 31 of the 34 references it cites were found in a registry or an index; none of the 15 sources its claims cite is checked for support yet.

Abstract

We present a novel rule-based system for automatic generation of factual questions from sentences, using semantic role labeling (SRL) as the main form of text analysis.The system is capable of generating both wh-questions and yes/no questions from the same semantic analysis.We present an extensive evaluation of the system and compare it to a recent neural network architecture for question generation.The SRL-based system outperforms the neural system in both average quality and variety of generated questions.

Overview

CoChat Summary

## TL;DR

The paper presents a rule-based open-domain question generation system that uses semantic role labeling to generate both wh-questions and yes/no questions from the same analysis. On 171 sentences, it outperformed an LSTM attention-based neural baseline in average quality, variety, and the share of questions judged potentially usable.

## Key findings

  • The SRL-based system generated questions for 165 of 171 sentences, producing 890 questions in total: 236 yes/no questions and 654 constituent questions.
  • Mean ratings favored the SRL system over the neural system on all three scales and on total score: SRL-YNQ 4.32/4.34/2.75/11.41, SRL-CQ 3.89/3.79/2.52/10.20, NN 3.75/2.61/1.65/8.01 for grammar/semantics/relevance/total.
  • The best SRL constituent question per sentence had a mean total rating of 12.2 versus 8.1 for the neural system’s questions; the difference was statistically significant (t-test, p < .0001).
  • Using the paper’s “potentially useful” criteria, 81% of SRL yes/no questions, 64% of SRL constituent questions, and 29% of neural questions had total rating ≥ 10; with the stricter per-scale threshold, the proportions were 71%, 50%, and 15%, respectively.

## Methods

  • Rule-based AQG pipeline: tokenization, sentence boundary detection, POS tagging with OpenNLP, verbal-group detection, semantic role labeling with SENNA, postprocessing, and question generation.
  • Data/sample for evaluation: five expository texts, yielding 171 declarative sentences (at least five words, no conditional clauses); the SRL and neural systems each generated questions for all 171 sentences, though the SRL system failed on 6 sentences and the neural system on 2.
  • Human evaluation: 1,060 generated questions were rated by two annotators on grammar, semantics, and relevance; analysis used average ratings, ANOVA with Bonferroni-adjusted pairwise contrasts, and a t-test for best-per-sentence constituent questions.

## Limitations

  • The paper states the SRL system falters on long and complicated sentences, especially those with subordinate clauses.
  • The neural baseline cannot generate yes/no questions because it is trained on SQuAD, which contains only constituent-style questions; the paper notes that beam-search hypotheses were not fully formed and only small variants of the best question.
  • Evident from the text: the evaluation is limited to 171 sentences from five expository texts, so the results are based on a relatively small and specific sample.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.18653/v1/w18-0530

Cites

31 of 31 references verified0 citation errors · 0 not checkable

Statistics

Recomputing p-values needs the paper’s full text.

Verification

Claims the paper makes

0 Valid · 1 Not valid · 8 Not checked · 51 Own results and interpretations

24 more claims

Create a free account to see all 60 claims and to verify the 15 references still open.

See all claims — free account
  • It may state two separate claims as one. 18
  • Probably not a substantial claim. 14

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Citations

Cites

Show all 31 references

Research works printed but not linked: 1

  1. Frank R. Palmer. 1987. The English Verb , 2nd edition. Longman, London, UK.

Cited by

See all 39

Metadata

Authors
Flor, M., & Riordan, B.
Affiliations
Educational Testing Service (US)
Venue
Proceedings of the Thirteenth Workshop on Innovative Use of NLP for Building Educational Applications
Published
2018
Pages
254-263
OpenAlex
W2807132464
Cited by
39
References
34
Citations by year
2026: 2 · 2025: 5 · 2024: 1 · 2023: 7 · 2022: 2 · 2021: 10 · 2020: 4 · 2019: 7 · 2018: 1
FWCI
2.25
Type
conference paper
Language
en
Open access
Licence
cc-by
Locations
Workshop on Innovative Use of NLP for Building Educational Applications (publishedVersion, cc-by)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
  • Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
  • Physical Sciences › Computer Science › Artificial Intelligence › Speech and dialogue systems
Keywords
automatic question generation, semantic role labeling, factual questions, wh-questions, yes-no questions

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-10-11, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-11 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: A Semantic Role-based Approach to Open-Domain Automatic Question Generation. https://app.cochat.ai/papers/10.18653/v1/w18-0530
The paper
Flor, M., & Riordan, B. (2018). A Semantic Role-based Approach to Open-Domain Automatic Question Generation. Proceedings of the Thirteenth Workshop on Innovative Use of NLP for Building Educational Applications. https://doi.org/10.18653/v1/w18-0530

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.