India

Glossing is the task of translating from a written language into a sequence of glosses, i.e., textual representations of signs from some sign language. While glossing is in principle &#39;just&#39; a machine translation (MT) task, sign languages still lack the large parallel corpora that exist for many written language pairs and underlie the development of dedicated MT systems. 
In this work, we demonstrate that glossing can be significantly improved through data augmentation. We fine-tune a Spanish transformer model both on a small dedicated corpus 3,000 Spanish–Mexican Sign Language (MSL) gloss sentence pairs, and on a corpus augmented with an English–American Sign Language (ASL) gloss corpus. We obtain the best results when we oversample from the ASL corpus by a factor of ~4, achieving a BLEU increase from 62 to 85 and a TER reduction from 44 to 20. This demonstrates the usefulness of combining corpora in low-resource glossing situations.

IJCNLP-AACL 2025

Low-Resource Sign Language Glossing Profits From Data Augmentation

mexican sign language

low resource language

data augmentation

### Welcome to IJCNLP-AACL 2025! 
 It is a great honor to host this joint conference in Mumbai, India, from December 20 to 24, 2025. The joint conferences of IJCNLP and AACL are organized with alternating leadership in the Asia-Pacific region. The event is run by the Asian Federation of Natural Language Processing (AFNLP) in odd years, and by AACL in even years, while it is organized solely by ACL when the annual ACL meeting is held in the region. This year, the conference is primarily organized by AFNLP. 
*Kentaro Inui
MBZUAI, UAE
General Chair, IJCNLP-AACL 2025* 
Read full message and download the Conference Handbook [**here**](https://drive.google.com/file/d/1UTwxkAqSqI-GAoJC3wE1zZt5VP1Y8GX0/view?usp=sharing).

The 14th IJCNLP & 4th AACL will be held in Mumbai, India from December 20th to December 24th, 2025.

Glossing is the task of translating from a written language into a sequence of glosses, i.e., textual representations of signs from some sign language. While glossing is in principle 'just' a machine translation (MT) task, sign languages still lack the large parallel corpora that exist for many written language pairs and underlie the development of dedicated MT systems. 
In this work, we demonstrate that glossing can be significantly improved through data augmentation. We fine-tune a Spanish transformer model both on a small dedicated corpus 3,000 Spanish–Mexican Sign Language (MSL) gloss sentence pairs, and on a corpus augmented with an English–American Sign Language (ASL) gloss corpus. We obtain the best results when we oversample from the ASL corpus by a factor of ~4, achieving a BLEU increase from 62 to 85 and a TER reduction from 44 to 20. This demonstrates the usefulness of combining corpora in low-resource glossing situations.

workshop paper

Sign language translation (SLT) is a challenging task due to the scarcity of labeled data and the heavy-tailed distribution of sign language vocabularies. In this paper, we explore a novel data augmentation approach for SLT: using a large language model (LLM) to generate paraphrases of the target language sentences in the training data. We experiment with a Transformer-based SLT model (Signformer) on three datasets spanning German, Greek, and Argentinian Sign Languages. For models trained with augmentation, we adopt a two-stage regime: pre-train on the LLM-augmented corpus and then fine-tune on the original, non-augmented training set. Our augmented training sets, expanded with GPT-4-generated paraphrases, yield mixed results. On a medium-scale German SL corpus (PHOENIX14T), LLM augmentation improves BLEU-4 from 9.56 to 10.33. In contrast, a small-vocabulary Greek SL dataset with a near-perfect baseline (94.38 BLEU) sees a slight drop to 92.22 BLEU, and a complex Argentinian SL corpus with a long-tail vocabulary distribution remains around 1.2 BLEU despite augmentation. We analyze these outcomes in relation to each dataset’s complexity and token frequency distribution, finding that LLM-based augmentation is more beneficial when the dataset contains a richer vocabulary and many infrequent tokens. To our knowledge, this work is the first to apply LLM paraphrasing to SLT, and we discuss these results with respect to prior data augmentation efforts in sign language translation.

Augmenting Sign Language Translation Datasets with Large Language Models

Sign languages are highly diverse across countries and regions, yet most Sign Language Translation (SLT) work remains monolingual. We explore a unified, multi-target SLT model trained jointly on four sign languages (German, Greek, Argentinian, Indian) using a standardized data layer. Our model operates on pose keypoints extracted with MediaPipe, yielding a lightweight and dataset-agnostic representation that is less sensitive to backgrounds, clothing, cameras, or signer identity while retaining motion and configuration cues. On RWTH-PHOENIX-Weather 2014T, Greek Sign Language Dataset, LSA-T, and ISLTranslate, naive joint training under a fully shared parameterization performs worse than monolingual baselines; however, a simple two-stage schedule: multilingual pre-training followed by a short language-specific fine-tuning, recovers and surpasses monolingual results on three datasets (PHOENIX14T: $+0.15$ BLEU-4; GSL: $+0.74$; ISL: $+0.10$) while narrowing the gap on the most challenging corpus (LSA-T: $-0.24$ vs.\ monolingual). Scores span from BLEU-4$\approx 1$ on open-domain news (LSA-T) to $>90$ on constrained curricula (GSL), highlighting the role of dataset complexity. We release our code to facilitate training and evaluation of multilingual SLT models.

Multilingual Sign Language Translation with Unified Datasets and Pose-Based Transformers

Fingerspelling enables signers to represent proper nouns and technical terms letter-by-letter using manual alphabets, yet remains severely under-resourced for Indian Sign Language (ISL). We present the first continuous fingerspelling dataset for ISL, extracted from the ISH News YouTube channel in which fingerspelling is accompanied by synchronized on-screen text cues. The dataset comprises 1,308 segments from 499 videos, totaling 70.85 minutes and 14,814 characters, with aligned video-text pairs capturing authentic coarticulation patterns. We validate dataset quality through annotation by a proficient ISL interpreter, achieving a 90.67% exact match rate for 150 samples. We further establish baseline recognition benchmarks using a ByT5-small encoder-decoder model, which attains 82.91% Character Error Rate after fine-tuning. This resource supports multiple downstream tasks including fingerspelling transcription, temporal localization, and sign generation. Although limited in scale, it represents one of the few available resources targeting continuous fingerspelling in ISL.

Continuous Fingerspelling Dataset for Indian Sign Language

Sign language translation (SLT) has witnessed rapid progress in the deep learning community across several sign languages, including German, American, British, and Italian. However, Indian Sign Language (ISL) remains relatively underexplored. Motivated by recent efforts to develop large-scale ISL resources, we investigate how existing SLT models perform on ISL data. Specifically, we evaluate three approaches: (i) training a transformer-based model, (ii) leveraging visual-language pretraining, and (iii) tuning a language model with pre-trained visual and motion representations. Unlike existing methods that primarily use raw video frames, we augment the model with optical flow maps to explicitly capture motion primitives, combined with a multi-scale feature extraction method for encoding spatial features (SpaMo-OF). Our approach achieves promising results, obtaining a BLEU-4 score of $8.58$ on the iSign test set, establishing a strong baseline for future ISL translation research. We will publicly release our code (upon acceptance of our manuscript).

Enhancing Indian Sign Language Translation via Motion-Aware Modeling

This paper presents a lightweight and efficient baseline for isolated Indian Sign Language (ISL) word recognition developed for
the WSLP-AACL-2025 Shared Task. We
propose a two-stage framework combining
skeletal landmark extraction via MediaPipe
Holistic with a Temporal Convolutional Network (TCN) for temporal sequence classification. The system processes pose-based
input sequences instead of raw video, significantly reducing computation and memory
costs. Trained on the WSLP-AACL-2025
dataset containing 4,398 isolated sign videos
across 4,361 word classes, our model achieves
54% top-1 and 78% top-5 accuracy.

Pose-Based Temporal Convolutional Networks for Isolated Indian Sign Language Word Recognition

This paper presents a cross-linguistic analysis of phonological similarity in sign languages using symbolic representations from the Hamburg Notation System (HamNoSys). We construct a dataset of 1000 signs each from British Sign Language (BSL), German Sign Language (DGS), French Sign Language (LSF), and Greek Sign Language (GSL), and compute pairwise phonological similarity using normalized edit distance over HamNoSys strings. Our analysis reveals both universal and language-specific patterns in handshape usage, movement dynamics, non-manual features, and spatial articulation. We explore intra and inter-language similarity distributions, phonological clustering, and co-occurrence structures across feature types. The findings offer insights into the structural organization of sign language phonology and highlight typological variation shaped by linguistic and cultural factors.

Cross-Linguistic Phonological Similarity Analysis in Sign Languages Using HamNoSys

Automatic Sign Language Recognition (ASLR) has emerged as a vital field for bridging the gap between deaf and hearing communities. However, the problem of sign-to-sign retrieval or detecting a specific sign within a sequence of continuous signs remains largely unexplored. We define this novel task as Sign Language Spotting. In this paper, we present a first step toward sign language retrieval by addressing the challenge of detecting the presence or absence of a query sign video within a sentence-level gloss or sign video. Unlike conventional approaches that rely on intermediate gloss recognition or text-based matching, we propose an end-to-end model that directly operates on pose keypoints extracted from sign videos. Our architecture employs an encoder-only backbone with a binary classification head to determine whether the query sign appears within the target sequence. By focusing on pose representations instead of raw RGB frames, our method significantly reduces computational cost and mitigates visual noise. We evaluate our approach on the Word Presence Prediction dataset from the WSLP 2025 shared task, achieving 61.88\% accuracy and 60.00\% F1-score. These results demonstrate the effectiveness of our pose-based framework for Sign Language Spotting, establishing a strong foundation for future research in automatic sign language retrieval and verification.

Pose-Based Sign Language Spotting via an End-to-End Encoder Architecture

Sign Language Translation (SLT) is a crucial technology for fostering communication accessibility for the Deaf and Hard-of-Hearing (DHH) community. A dominant approach in SLT involves a two-stage pipeline: first, transcribing video to sign language glosses, and then translating these glosses into natural text. This second stage, gloss-to-text translation, is a challenging, low-resource machine translation task due to data scarcity and significant syntactic divergence. While prior work has often relied on training translation models from scratch, we show that fine-tuning large, pre-trained language models (PLMs) offers a more effective and data-efficient paradigm. In this work, we conduct a comprehensive bidirectional evaluation of several PLMs (T5, Flan-T5, mBART, and Llama) on this task. We use a collection of popular SLT datasets (RWTH-PHOENIX-14T, SIGNUM, and ASLG-PC12) and evaluate performance using standard machine translation metrics. Our results show that fine-tuned PLMs consistently and significantly outperform Transformer models trained from scratch, establishing new state-of-the-art results. Crucially, our bidirectional analysis reveals a significant performance gap, with Text-to-Gloss translation posing a greater challenge than Gloss-to-Text. We conclude that leveraging the linguistic knowledge of pre-trained models is a superior strategy for gloss translation and provides a more practical foundation for building robust, real-world SLT systems.

Finetuning Pre-trained Language Models for Bidirectional Sign Language Gloss to Text Translation

Quantum Natural Language Processing (QNLP) has emerged as a novel paradigm that leverages the principles of quantum mechanics to address fundamental challenges in language modeling, particularly in capturing compositional meaning. This survey charts the evolution of QNLP, from its theoretical foundations in the Distributional Compositional Categorical (DisCoCat) framework to its modern implementation on Noisy Intermediate-Scale Quantum (NISQ) hardware. We review the primary architectural approaches, including variational quantum circuits and tensor networks, and summarize the growing body of empirical work in tasks such as text classification, sentence similarity, and question answering. A recurring finding is the potential for QNLP models to achieve competitive performance with significantly fewer parameters than their classical counterparts. However, the field is critically constrained by the limitations of NISQ-era hardware. We conclude by discussing these challenges and outlining the future trajectory towards achieving a demonstrable quantum advantage and building more interpretable, efficient language models.

A Survey of Quantum Natural Language Processing: From Compositional Models to NISQ-Era Empiricism

Legal document analysis is pivotal in modern judicial systems, particularly for case retrieval, classification, and recommendation tasks. Graph neural networks (GNNs) have revolutionized legal use cases by enabling the efficient analysis of complex relationships. Although existing legal citation network datasets have significantly advanced research in this domain, the lack of large-scale open-source datasets tailored to the Indian judicial system has limited progress. To address this gap, we present the Indian Legal Citation Network (LeCNet) - the first open-source benchmark dataset for the link prediction task (missing citation recommendation) in the Indian judicial context. The dataset has been created by extracting information from the original judgments. LeCNet comprises 26,308 nodes representing case judgments and 67,108 edges representing citation relationships between the case nodes. Each node is described with rich features of document embeddings that incorporate contextual information from the case documents. Baseline experiments using various machine learning models were conducted for dataset validation. The Mean Reciprocal Rank (MRR) metric is used for model evaluation. The results obtained demonstrate the utility of the LeCNet dataset, highlighting the advantages of graph-based representations over purely textual models.

Downloads

Next from IJCNLP-AACL 2025

Augmenting Sign Language Translation Datasets with Large Language Models

Similar lecture

Improving Sign Language Understanding with a Multi-Stream Masked Autoencoder Trained on ASL Videos

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from IJCNLP-AACL 2025

Augmenting Sign Language Translation Datasets with Large Language Models

Similar lecture

Improving Sign Language Understanding with a Multi-Stream Masked Autoencoder Trained on ASL Videos

Downloads