Papers by Stephen Wan
Detecting Online Community Practices with Large Language Models: A Case Study of Pro-Ukrainian Publics on Twitter (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for identifying practices within social media are not yet available. |
| Approach: | They propose a methodological workflow for computational identification of such practices within social media texts by using open-source models and OpenAI’s large language models. |
| Outcome: | The proposed method improves accuracy and supports context-sensitive moderation and advancing the understanding of online community dynamics. |
A Position Paper on the Automatic Generation of Machine Learning Leaderboards (2025.emnlp-main)
Copied to clipboard
| Challenge: | Automated leaderboard generation is a tool for comparing prior work with a tabular overview of experimental results. |
| Approach: | They propose an automatic leaderboard generation framework to standardise how the task is defined. |
| Outcome: | The proposed framework standardises how the ALG task is defined and proposes new directions . the proposed framework includes recommendations for datasets and metrics that promote fair evaluation . |
What Causes the Failure of Explicit to Implicit Discourse Relation Recognition? (2024.naacl-long)
Copied to clipboard
| Challenge: | Prior work claimed that explicit classifiers perform poorly in implicit scenarios . a label shift occurs after connectives are removed, but no empirical evidence supports this claim . |
| Approach: | They propose to prove that the discourse relations expressed by some explicit instances will change when connectives disappear. |
| Outcome: | The proposed methods outperform strong baselines on PDTB 2.0, PDTT 3.0, and the GUM dataset. |
How to Best Use Syntax in Semantic Role Labelling (P19-1)
Copied to clipboard
| Challenge: | Existing studies on integrating external information into NLP tasks focus on word-level shallow features such as POS or chunk tags. |
| Approach: | They propose to integrate syntactic information into a neural ELMo-based SRL sequence labelling model by using a constituency representation as input features. |
| Outcome: | The proposed approach improves performance on the in-domain CoNLL’05 and CoNll’12 benchmarks. |
Integrating Lexical Information into Entity Neighbourhood Representations for Relation Prediction (2021.naacl-main)
Copied to clipboard
| Challenge: | Existing methods to predict knowledge base relations are limited by maintenance costs and text-based formats. |
| Approach: | They propose a system that can extend relational database tables with information extracted from a document corpus. |
| Outcome: | The proposed system outperforms existing methods by incorporating embeddings of text-based representations of the entities and relations. |
Mention Flags (MF): Constraining Transformer-based Text Generators (2021.acl-long)
Copied to clipboard
| Challenge: | Constrained decoding algorithms produce hypotheses satisfying all constraints, but they are computationally expensive and can lower the generated text quality. |
| Approach: | They propose a Mention Flag mechanism which traces whether lexical constraints are satisfied in outputs of an S2S decoder. |
| Outcome: | The proposed models maintain higher constraint satisfaction and text quality than baseline models and other constrained decoding algorithms. |
MetaLead: A Comprehensive Human-Curated Leaderboard Dataset for Transparent Reporting of Machine Learning Experiments (2026.eacl-long)
Copied to clipboard
| Challenge: | Existing leaderboards capture only the best results from each paper and have limited metadata. |
| Approach: | They propose to create a fully human-annotated ML Leaderboard dataset that captures all experimental results and contains extra metadata. |
| Outcome: | The MetaLead dataset captures all experimental results and contains extra metadata for cross-domain evaluation. |
ECOL-R: Encouraging Copying in Novel Object Captioning with Reinforcement Learning (2021.eacl-main)
Copied to clipboard
| Challenge: | Novel Object Captioning is a zero-shot Image Caption task requiring describing objects not seen in the training captions, but for which information is available from external object detectors. |
| Approach: | They propose a novel captioning model that encourages copying of object labels with reinforcement learning that encourage a copy-augmented transformer model to accurately describe the object labels. |
| Outcome: | The proposed model sets new state-of-the-art on the nocaps and held-out COCO benchmarks. |
Using a Human-AI Teaming Approach to Create and Curate Scientific Datasets with the SciLire System (2026.eacl-demo)
Copied to clipboard
Necva Bölücü, Jessica Irons, Changhyun Lee, Brian Jin, Maciej Rybinski, Huichen Yang, Andreas Duenser, Stephen Wan
| Challenge: | rapid growth of scientific literature has made manual extraction of structured knowledge increasingly impractical. |
| Approach: | They propose a system for creating datasets from scientific literature that integrates human-AI teaming principles and iterative workflows. |
| Outcome: | The proposed system improves extraction fidelity and facilitates efficient dataset creation. |
impact of sample selection on in-context learning for entity extraction from scientific writing (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Prompt-based use of Large Language Models is becoming popular . specialized domains such as entity extraction are expensive to annotate . |
| Approach: | They propose to use a prompt set-up to provide training examples along with the inference request. |
| Outcome: | The proposed methods improve on a fully supervised transformer-based baseline. |