Papers by Holger Schwenk

18 papers
Filtering and Mining Parallel Data in a Joint Multilingual Space (P18-2)

Copied to clipboard

Challenge: Using a cosine distance in a joint multilingual sentence embedding, we filter out noisy parallel data and mine for bitexts in large news collections.
Approach: They propose to learn a joint multilingual sentence embedding and use the distance between sentences in different languages to filter noisy parallel data and to mine for parallel data in large monolingual texts.
Outcome: The proposed approach improves a competitive baseline on the WMT'14 task by 0.3 BLEU by filtering out 25% of the training data.
WikiMatrix: Mining 135M Parallel Sentences in 1620 Language Pairs from Wikipedia (2021.eacl-main)

Copied to clipboard

Challenge: a new approach to extract parallel sentences from Wikipedia articles is proposed . the approach is based on multilingual sentence embeddings, but does not limit it to English .
Approach: They propose to automatically extract parallel sentences from Wikipedia articles in 96 languages . they train neural MT baseline systems on the mined data and evaluate them on the TED corpus .
Outcome: The proposed approach extracts parallel sentences from Wikipedia articles in 96 languages . the extracted sentences achieve strong BLEU scores for many language pairs .
Margin-based Parallel Corpus Mining with Multilingual Sentence Embeddings (P19-1)

Copied to clipboard

Challenge: Traditional parallel corpus mining methods focus on the textual content instead of the size and quality of training data.
Approach: They propose a method for machine translation based on multilingual sentence embeddings.
Outcome: The proposed method outperforms the best published methods on the BUCC mining task and the UN reconstruction task by more than 10 F1 and 30 precision points.
SpeechMatrix: A Large-Scale Mined Corpus of Multilingual Speech-to-Speech Translations (2023.acl-long)

Copied to clipboard

Challenge: SpeechMatrix is a large-scale multilingual corpus of speech-to-speech translations mined from real speech of European Parliament recordings.
Approach: They present a large-scale multilingual corpus of speech-to-speech translations mined from real speech of European Parliament recordings.
Outcome: The proposed model can train bilingual models on 136 language pairs with 418 thousand hours of speech.
CCMatrix: Mining Billions of High-Quality Parallel Sentences on the Web (2021.acl-long)

Copied to clipboard

Challenge: Using a curated common crawl corpus, we were able to mine 10.8 billion parallel sentences out of which only 2.9 billions are aligned with English.
Approach: They use 32 snapshots of a curated common crawl corpus totaling 71 billion unique sentences to mine 10.8 billion parallel sentences out of which only 2.9 billions are aligned with English.
Outcome: The proposed system outperforms the best single systems on the WMT’19 test set for English-German/Russian/Chinese and outperformed the best submission at the 2020 WAT workshop.
MLQA: Evaluating Cross-lingual Extractive Question Answering (2020.acl-main)

Copied to clipboard

Challenge: Question answering (QA) models have shown rapid progress enabled by the availability of large, high-quality benchmark datasets.
Approach: They present a multi-way aligned extractive QA evaluation benchmark in 7 languages . they evaluate state-of-the-art cross-lingual models and machine-translation-based baselines .
Outcome: The proposed model is based on MLQA, which has over 12K instances in english and 5K in each other language.
A Corpus for Multilingual Document Classification in Eight Languages (L18-1)

Copied to clipboard

Challenge: a subset of the Reuters corpus volume 2 is used to evaluate cross-lingual document classification . current best practice is to evaluate document classification on resources in one language and transfer it to another without additional resources.
Approach: They propose to use a subset of the Reuters corpus to evaluate cross-lingual document classification . they propose to add Italian, Russian, Japanese and Chinese to the subset .
Outcome: The proposed subset of the Reuters corpus has balanced class priors for eight languages.
Speech-to-Speech Translation for a Real-world Unwritten Language (2023.findings-acl)

Copied to clipboard

Challenge: a new study examines speech-to-speech translation (S2ST) that translates speech from one language into another . the research area for unwritten languages remains a research area with little exploration due to the lack of training data.
Approach: They propose a system that translates speech from one language into another . they use Taiwanese Hokkien as an example of an unwritten language .
Outcome: The proposed system can be used to train models in languages without standard writing systems.
Textless Speech-to-Speech Translation on Real Data (2022.naacl-main)

Copied to clipboard

Challenge: Existing text-based speech-to-speech translation systems rely on cascaded approach . text-to text translation systems require text generation and a single input to generate output .
Approach: They propose a textless speech-to-speech translation system that can translate speech from one language into another without the need of text data.
Outcome: The proposed system can translate speech from one language into another without text data.
Bitext Mining Using Distilled Sentence Representations for Low-Resource Languages (2022.findings-emnlp)

Copied to clipboard

Challenge: a new study aims to extend multilingual representation learning beyond the hundred most frequent languages . current work on multilingual sentence representations has focused on training one model which handles all languages of interest .
Approach: They propose a teacher-student approach to extend existing monolingual sentence embedding space to new languages.
Outcome: The proposed model outperforms the original LASER encoder in 44 African languages . the model can be used to train multiple languages and learn new languages if they have the same training data .
Aligning Speech Segments Beyond Pure Semantics (2024.findings-acl)

Copied to clipboard

Challenge: Existing speech-to-speech parallel data is scarce and expensive to create from scratch.
Approach: They propose an algorithm which automatically aligns pairs of speech segments aligned in meaning and expressivity.
Outcome: The proposed algorithm outperforms semantic-focused approaches on content translation quality.
xSIM++: An Improved Proxy to Bitext Mining Performance for Low-Resource Languages (2023.acl-short)

Copied to clipboard

Challenge: xsim++ provides a reliable proxy for bitext mining without expensive pipelines.
Approach: They propose a new proxy proxy based on similarity in a multilingual embedding space . they validate this proxy by running a significant number of bitext mining experiments for a set of low-resource languages and then train NMT systems on the mined data.
Outcome: The proposed proxy improves on xsim++ and trains on the mined data.
LCFO: Long Context and Long Form Output Dataset and Benchmarking (2025.findings-acl)

Copied to clipboard

Challenge: Using long text outputs to evaluate progress in summarization and summary expansion tasks is challenging.
Approach: They propose a framework for assessing gradual summarization and summary expansion capabilities across diverse domains.
Outcome: The proposed framework provides alignments between specific QA pairs and corresponding summaries in 7 domains.
XNLI: Evaluating Cross-lingual Sentence Representations (D18-1)

Copied to clipboard

Challenge: State-of-the-art natural language processing systems rely on annotated data to learn competent models.
Approach: They extend the development and test sets of the Multi-Genre Natural Language Inference Corpus to 14 languages, including Swahili and Urdu.
Outcome: The proposed evaluation set extends the development and test sets of the Multi-Genre Natural Language Inference Corpus (MultiNLI) to 14 languages including low-resource languages such as Swahili and Urdu.
Multilingual Representation Distillation with Contrastive Learning (2023.eacl-main)

Copied to clipboard

Challenge: Contextual representations from large pretrained language models encode semantic information from two or more languages.
Approach: They integrate contrastive learning into multilingual representation distillation and use it for quality estimation of parallel sentences.
Outcome: The proposed model outperforms existing models with similarity searches and filtering tasks across low-resource languages.
stopes - Modular Machine Translation Pipelines (2022.emnlp-demos)

Copied to clipboard

Challenge: Neural machine translation is a natural language deep learning application that needs data to be trained.
Approach: They describe a framework that empowers scalability and versatility for research use cases.
Outcome: The proposed framework empowers scalability and versatility for research use cases.
BLASER: A Text-Free Speech-to-Speech Translation Evaluation Metric (2023.acl-long)

Copied to clipboard

Challenge: End-to-End speech-to speech translation is generally evaluated with text-based metrics . this means generated speech has to be automatically transcribed, making the evaluation dependent on ASR systems.
Approach: They propose a text-free evaluation metric for end-to-end speech-tospeech translation, named BLASER, to avoid the dependency on automatic speech recognition systems.
Outcome: The proposed metric avoids the dependency on automatic speech recognition systems by encoding generated speech segments into a shared embedding space.
T-Modules: Translation Modules for Zero-Shot Cross-Modal Machine Translation (2022.emnlp-main)

Copied to clipboard

Challenge: Existing approaches to perform zero-shot cross-modal transfer between speech and text are limited to a very small number of language pairs.
Approach: They propose a method to perform zero-shot cross-modal transfer between speech and text for translation tasks by using a speech decoder.
Outcome: The proposed model significantly improves state-of-the-art for zero-shot speech translation on Must-C.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations