Papers by Jan Buys
Neural Machine Translation between Low-Resource Languages with Synthetic Pivoting (2024.lrec-main)
Copied to clipboard
| Challenge: | Pivot-based neural machine translation systems overcome data scarcity by including a high-resource pivot language in the process of translating between low-resourced languages. |
| Approach: | They propose a novel approach to pivot-based translation in which pivot sentences are generated synthetically from both the source and target languages. |
| Outcome: | The proposed approach improves pivot-based systems translating between low-resource Southern African languages by up to 5.6 BLEU points after fine-tuning. |
Discourse Understanding and Factual Consistency in Abstractive Summarization (2021.eacl-main)
Copied to clipboard
Saadia Gabriel, Antoine Bosselut, Jeff Da, Ari Holtzman, Jan Buys, Kyle Lo, Asli Celikyilmaz, Yejin Choi
| Challenge: | Existing abstractive summarization models often hallucinate information or generate factually incorrect summaries. |
| Approach: | They propose a general framework for abstractive summarization with factual consistency and distinct modeling of the narrative flow in an output summary. |
| Outcome: | The proposed framework generates abstracts with factual consistency and coherence significantly better than baselines. |
Subword Segmental Machine Translation: Unifying Segmentation and Target Sentence Generation (2023.findings-acl)
Copied to clipboard
| Challenge: | Subword segmenters are used in neural machine translation, but are not used in high-resource settings. |
| Approach: | They propose a subword segmental machine translation (SSMT) that unifies subword and MT in a single trainable model. |
| Outcome: | The proposed model improves chrF scores for morphologically rich agglutinative languages and is more robust on a test set constructed for evaluating morphology generalisations. |
Triples-to-isiXhosa (T2X): Addressing the Challenges of Low-Resource Agglutinative Data-to-Text Generation (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing data-to-text models are designed for the linguistic typology of English, but they are not suitable for low-resource languages. |
| Approach: | They propose a new dataset based on a subset of WebNLG that is agglutinative and low-resource data-to-text. |
| Outcome: | The proposed model outperforms existing models for isiXhosa and Finnish and fine-tunes machine translation models as the best method overall. |
BottleSum: Unsupervised and Self-supervised Sentence Summarization using the Information Bottleneck Principle (D19-1)
Copied to clipboard
| Challenge: | Existing approaches to extractive and abstractive summarization rely on large-scale parallel corpora of input text and output summaries for direct supervision. |
| Approach: | They propose an unsupervised approach to sentence summarization using the Information Bottleneck principle. |
| Outcome: | The proposed method outperforms unsupervised models on automatic metrics and human evaluation along multiple attributes. |
Multi-Hall-SA: A Cross-lingual Benchmark for Multi-Type Hallucination Detection in Low-Resource South African Languages (2026.findings-eacl)
Copied to clipboard
| Challenge: | Large Language Models generate false or unsupported information, which can be difficult to detect in low-resource languages. |
| Approach: | They propose a cross-lingual benchmark for hallucination detection spanning English and South African languages. |
| Outcome: | The proposed model detects 23.6% fewer hallucinations in South African languages compared to English . human validation confirms the quality and cross-lingual alignment of the model . |
A Systematic Analysis of Subwords and Cross-Lingual Transfer in Multilingual Translation (2024.findings-naacl)
Copied to clipboard
| Challenge: | Multilingual modelling can improve machine translation for low-resource languages, partly through shared subword representations. |
| Approach: | They propose to use subword regularisation to promote synergy and BPE to facilitate cross-lingual transfer. |
| Outcome: | The proposed methods promote synergy and prevent interference across different linguistic typologies. |
Subword Segmental Language Modelling for Nguni Languages (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Subword segmentation is a standard practice in NLP, but is viewed as a preprocessing step for low-resource languages with complex morphologies. |
| Approach: | They propose a subword segmental language model that learns how to segment words while being trained for autoregressive language modelling. |
| Outcome: | The proposed model outperforms existing models on unsupervised morphological segmentation and outperfies standard subword segmenters on all 4 languages. |
Cross-Lingual Knowledge Projection and Knowledge Enhancement for Zero-Shot Question Answering in Low-Resource Languages (2025.coling-main)
Copied to clipboard
| Challenge: | Knowledge bases (KBs) in low-resource languages are often incomplete, restricting the ability to do zero-shot question answering using multilingual language models. |
| Approach: | They propose a novel cross-lingual mapping technique which improves word alignments extracted from parallel English-LRL text by combining lexical alignment, named entity recognition, and semantic alignment. |
| Outcome: | The proposed approach improves zero-shot question answering accuracy by up to 17% compared to baselines without KB access. |
Multipath parsing in the brain (2024.acl-long)
Copied to clipboard
| Challenge: | a major unsolved problem in computational psycholinguistics is determining whether human comprehension considers a single analysis path 1 at a time. |
| Approach: | They compare syntactic surprisal from a state-of-the-art dependency parser with fMRI data . they find evidence for multipath parsing in English and Chinese data based on fm data a major unsolved problem in computational psycholinguistics is determining whether human sentence comprehension considers a single analysis path 1 at a time . |
| Outcome: | The proposed model shows that human parsing is multipath, with a higher r2 increase for multipath surprisal than single-path surpresal. |
RepGraph: Visualising and Analysing Meaning Representation Graphs (2021.emnlp-demo)
Copied to clipboard
| Challenge: | Graph-based meaning representations provide rich semantic annotations, but visualising them clearly is more challenging than for fully lexicalized representations. |
| Approach: | They propose to use RepGraph to visualise, manipulate and analyse semantically parsed graph data in a JSON-based serialisation format. |
| Outcome: | The proposed visualisation and analysis tool supports DMRS, EDS, PTG, UCCA, and AMR semantic frameworks. |
Generic Overgeneralization in Pre-trained Language Models (2022.coling-1)
Copied to clipboard
| Challenge: | Generic statements such as "ducks lay eggs" are perceived as false universally . however, universally quantified statements such "all tigers have stripes" should be perceived as true . |
| Approach: | They investigate the generic overgeneralization effect in pre-trained language models . they show that pre-trainers tend to treat quantified generic statements as if they were true . |
| Outcome: | The proposed model reduces, but does not eliminate, generic overgeneralization bias . the model can be used to inject factual knowledge about kinds into pre-trained models . |
Learning to Write with Cooperative Discriminators (P18-1)
Copied to clipboard
| Challenge: | Despite their local fluency, long-form text generated from RNNs is often generic, repetitive, and even self-contradictory. |
| Approach: | They propose a unified learning framework that can guide a base RNN generator towards more globally coherent generations by combining discriminators with a composite decoding objective. |
| Outcome: | The proposed framework can guide a base RNN generator towards more globally coherent generations by combining discriminators with the base RRN generator through a composite decoding objective. |
Policy-based Reinforcement Learning for Generalisation in Interactive Text-based Environments (2023.eacl-main)
Copied to clipboard
| Challenge: | Text-based environments allow RL agents to learn to converse and perform interactive tasks through natural language. |
| Approach: | They propose to switch from a value-based update method to a policy-based one within text-based environments and evaluate it on Coin Collector and Question Answering with interactive text (QAit). |
| Outcome: | The proposed policy-based agent is more generalised than value-based methods in two text-based environments designed to test zero-shot performance. |
Benchmarking Hierarchical Script Knowledge (N19-1)
Copied to clipboard
| Challenge: | Understanding procedural language requires reasoning about hierarchical and temporal relations between events. |
| Approach: | They propose a hierarchical script learning dataset and a cloze task to match video captions with missing procedural details. |
| Outcome: | The proposed model matches video captions with missing procedural details to find out if they can understand the language. |
Neural Syntactic Generative Models with Exact Marginalization (N18-1)
Copied to clipboard
| Challenge: | Recent models have added structure to recurrent neural networks at the cost of giving up exact inference, or using soft structure instead of latent variables. |
| Approach: | They propose a syntactic generative model with exact marginalization that supports dependency parsing and language modeling. |
| Outcome: | The proposed models achieve state-of-the-art for supervised dependency parsing and language modeling. |
Neural Text Generation from Rich Semantic Representations (N19-1)
Copied to clipboard
| Challenge: | 2 is a neural model that maps a linearization of Dependency MRS to text . 1 is based on a BLEU score of 66.11 when trained on gold data . |
| Approach: | They propose to use Minimal Recursion Semantics to generate high-quality text from structured representations. |
| Outcome: | The proposed model achieves a BLEU score of 77.17 on the full test set and 83.37 on the subset of test data most closely matching the silver data domain. |
NGLUEni: Benchmarking and Adapting Pretrained Language Models for Nguni Languages (2024.lrec-main)
Copied to clipboard
| Challenge: | Nguni languages have over 20 million home language speakers in South Africa . there has been considerable growth in the datasets for these languages, but no analysis of the performance of NLP models for these language has been reported across languages and tasks. |
| Approach: | They compile publicly available datasets for natural language understanding and generation, spanning 6 tasks and 11 datasets. |
| Outcome: | The proposed models outperform existing models and large-scale adapted models on cross-lingual transfer and machine translation. |