In this paper, we explore the utility of Translationese as synthetic data created using machine translation for pre-training language models (LMs) for low-resource languages (LRLs). Our simple methodology consists of translating large amounts of web-crawled monolingual documents (clean) into the LRLs, followed by filtering the translated documents using tiny LMs trained on small but clean LRL data. Taking the case of Indian languages, we pre-train LMs from scratch with 28M and 85M parameters, and then fine-tune them for 5 downstream natural language understanding (NLU) and 4 generative (NLG) tasks. We observe that pre-training on filtered synthetic data leads to relative performance drops of only 0.87% for NLU and 2.35% for NLG, compared to pre-training on clean data, and this gap further diminishes upon the inclusion of a small amount of clean data. We also study the impact of synthetic data filtering and the choice of source language for synthetic data generation. Furthermore, evaluating continually pre-trained larger models like Gemma-2B and Llama-3-8B in few-shot settings, we observe that using synthetic data is competitive with using clean data. Our findings suggest that synthetic data shows promise for bridging the pre-training gap between English and LRLs.

Contemporary deep learning models effectively handle languages with diverse morphology despite not being directly integrated into them. Morphology and word order are closely linked, with the latter incorporated into transformer-based models through positional encodings. This prompts a fundamental inquiry: Is there a correlation between the morphological complexity of a language and the utilization of positional encoding in pre-trained language models? In pursuit of an answer, we present the first study addressing this question, encompassing 22 languages and 5 downstream tasks. Our findings reveal that the importance of positional encoding diminishes with increasing morphological complexity in languages. Our study motivates the need for a deeper understanding of positional encoding, augmenting them to better reflect the different languages under consideration.

Creoles represent an under-explored and marginalized group of languages, with few available resources for NLP research. While the genealogical ties between Creoles and a number of highly resourced languages imply a significant potential for transfer learning, this potential is hampered due to this lack of annotated data. In this work we present CreoleVal, a collection of benchmark datasets spanning 8 different NLP tasks, covering up to 28 Creole languages; it is an aggregate of novel development datasets for reading comprehension relation classification, and machine translation for Creoles, in addition to a practical gateway to a handful of preexisting benchmarks. For each benchmark, we conduct baseline experiments in a zero-shot setting in order to further ascertain the capabilities and limitations of transfer learning for Creoles. Ultimately, we see CreoleVal as an opportunity to empower research on Creoles in NLP and computational linguistics, and in general, a step towards more equitable language technology around the globe.

Recent interest has surged in employing Large Language Models (LLMs) for machine translation (MT) via in-context learning (ICL) (Vilar et al., 2023). Most prior studies primarily focus on optimizing translation quality, with limited attention to understanding the specific aspects of ICL that influence the said quality. To this end, we perform the first of its kind, exhaustive study of in-context learning for machine translation (MT). We first establish that ICL is primarily example-driven and not instruction-driven. Following this, we conduct an extensive exploration of various aspects of the examples to understand their influence on downstream performance. Our analysis includes factors such as quality and quantity of demonstrations, spatial proximity, and source versus target originality. Further, we also investigate challenging scenarios involving indirectness and misalignment of examples to understand the limits of ICL. While we establish the significance of the quality of the target distribution over the source distribution of demonstrations, we further observe that perturbations sometimes act as regularizers, resulting in performance improvements. Surprisingly, ICL does not necessitate examples from the same task, and a related task with the same target distribution proves sufficient. We hope that our study acts as a guiding resource for considerations in utilizing ICL for MT. Our code is available on https://github.com/PranjalChitale/in-context-mt-analysis.

LLMs have demonstrated remarkable capability for understanding semantics, but their understanding of pragmatics is not well studied. To this end, we release a Pragmatics Understanding Benchmark (PUB) dataset consisting of fourteen tasks in four pragmatics phenomena, namely; Implicature, Presupposition, Reference, and Deixis. We curate high-quality test sets for each task, consisting of Multiple Choice Question Answers (MCQA). PUB includes a total of 28k data points, 6.1k are newly annotated. We evaluate nine models varying in the number of parameters and type of training. Our study reveals several key observations about the pragmatic capabilities of LLMs: 1. chat-fine-tuning strongly benefits smaller models, 2. large base models are competitive with their chat-fine-tuned counterparts, 3. there is a huge variance in performance across different pragmatics phenomena, and 4. a noticeable performance gap between human capabilities and model capabilities. We hope that PUB will enable comprehensive evaluation of LLM’s pragmatic reasoning capabilities.

Language Models (LMs) excel in natural language processing tasks for English but show reduced performance in most other languages. This problem is commonly tackled by continually pre-training and fine-tuning these models for said languages. A significant issue in this process is the limited vocabulary coverage in the original model’s tokenizer, leading to inadequate representation of new languages and necessitating an expansion of the tokenizer. The initialization of the embeddings corresponding to new vocabulary items presents a further challenge. Current strategies require cross-lingual embeddings and lack a solid theoretical foundation as well as comparisons with strong baselines. In this paper, we first establish theoretically that initializing within the convex hull of existing embeddings is a good initialization, followed by a novel but simple approach, Constrained Word2Vec (CW2V), which does not require cross-lingual embeddings. Our study evaluates different initialization methods for expanding RoBERTa and LLaMA 2 across four languages and five tasks. The results show that CW2V performs equally well or even better than more advanced techniques. Additionally, simpler approaches like multivariate initialization perform on par with these advanced methods indicating that efficient large-scale multilingual continued pretraining can be achieved even with simpler initialization methods.

Recent works have shown that prompting large language models (LLMs) is effective for translation with markup where LLMs can simultaneously transfer markup tags while ensuring that the content, both inside and outside tag pairs is correctly translated. However, these works make a rather unrealistic assumption of the existence of high-quality parallel sentences with markup for prompting. Furthermore, the impact of instruction fine-tuning (IFT) in this setting is unknown. In this paper, we provide a study, the first of its kind, focusing on the effectiveness of synthetically created markup data and IFT for translation with markup using LLMs. We focus on translation from English to five European languages, German, French, Dutch, Finnish and Russian, where we show that regardless of few-shot prompting or IFT, synthetic data created via word alignments, while leading to inferior markup transfer compared to using original data with markups, does not negatively impact the translation quality. Furthermore, IFT mainly impacts the translation quality compared to few-shot prompting and has slightly better markup transfer capabilities than the latter. We hope our work will help practitioners make effective decisions on modeling choices for LLM based translation with markup.

Subword regularized models leverage multiple subword tokenizations of one target sentence during training. However, selecting one tokenization during inference leads to the underutilization of knowledge learned about multiple tokenizations.We propose the SubMerge algorithm to rescue the ignored Subword tokenizations through merging equivalent ones during inference.SubMerge is a nested search algorithm where the outer beam search treats the word as the minimal unit, and the inner beam search provides a list of word candidates and their probabilities, merging equivalent subword tokenizations. SubMerge estimates the probability of the next word more precisely, providing better guidance during inference.Experimental results on six low-resource to high-resource machine translation datasets show that SubMerge utilizes a greater proportion of a model’s probability weight during decoding (lower word perplexities for hypotheses). It also improves BLEU and chrF++ scores for many translation directions, most reliably for low-resource scenarios. We investigate the effect of different beam sizes, training set sizes, dropout rates, and whether it is effective on non-regularized models.

A majority of language technologies are tailored for a small number of high-resource languages, while relatively many low-resource languages are neglected. One such group, Creole languages, have long been marginalized in academic study, though their speakers could benefit from machine translation (MT). These languages are predominantly used in much of Latin America, Africa and the Caribbean. We present the largest cumulative dataset to date for Creole language MT, including 14.5M unique Creole sentences with parallel translations—11.6M of which we release publicly, and the largest bitexts gathered to date for 41 languages—the first ever for 21. In addition, we provide MT models supporting all 41 Creole languages in 172 translation directions. Given our diverse dataset, we produce a model for Creole language MT exposed to more genre diversity then ever before, which outperforms a genre-specific Creole MT model on its own benchmark for 23 of 34 translation directions.

Parallel data is difficult to obtain for low-resource languages in machine translation tasks, making it crucial to leverage monolingual linguistic features as auxiliary information. This article introduces a novel integration of hypernym features into the model by combining learnable hypernym embeddings with word embeddings, providing semantic information. Experimental results based on bilingual and multilingual models showed that: (1) incorporating hypernyms improves translation quality in low-resource settings, yielding +1.7 BLEU scores for bilingual models, (2) the hypernym feature demonstrates efficacy both in isolation and in conjunction with syntactic features, and (3) the performance is influenced by the choice of feature combination operators and hypernym-path hyperparameters.

This paper presents the NICT’s submission for the IWSLT 2024 Indic track, focusing on three speech-to-text (ST) translation directions: English to Hindi, Bengali, and Tamil. We aim to enhance translation quality in this low-resource scenario by integrating state-of-the-art pre-trained automated speech recognition (ASR) and text-to-text machine translation (MT) models. Our cascade system incorporates a Whisper model fine-tuned for ASR and an IndicTrans2 model fine-tuned for MT. Additionally, we propose an end-to-end system that combines a Whisper model for speech-to-text conversion with knowledge distilled from an IndicTrans2 MT model. We first fine-tune the IndicTrans2 model to generate pseudo data in Indic languages. This pseudo data, along with the original English speech data, is then used to fine-tune the Whisper model. Experimental results show that the cascaded system achieved a BLEU score of 51.0, outperforming the end-to-end model, which scored 19.1 BLEU. Moreover, the analysis indicates that applying knowledge distillation from the IndicTrans2 model to the end-to-end ST model improves the translation quality by about 0.7 BLEU.

This paper presents the findings of the WMT 2024’s MultiIndic22MT Shared Task, focusing on Machine Translation (MT) of 22 Indian Languages. In this task, we challenged participants with building MT systems which could translate between any or all of 22 Indian languages in the 8th schedule of the Indian constitution and English. For evaluation, we focused on automatic metrics, namely, chrF, chrF++ and BLEU.

While Marathi is considered as a low- to middle-resource language, its 42 dialects have mostly been ignored, mainly because these dialects are mostly spoken and rarely written, making them extremely low-resource. In this paper we explore the machine translation (MT) of Kadodi, also known as Samvedi, which is a dialect of Marathi. We first discuss the Kadodi dialect, highlighting the differences from the standard dialect, followed by presenting a manually curated dataset called Suman consisting of a trilingual Kadodi-Marathi-English dictionary of 949 entries and 942 simple sentence triples and idioms created by native Kadodi speakers. We then evaluate 3 existing large language models (LLMs) supporting Marathi, namely Gemma-2-9b, Sarvam-2b-0.5 and LLaMa-3.1-8b, in few-shot prompting style to determine their efficacy for translation involving Kadodi. We observe that these models exhibit rather lackluster performance in handling Kadodi even for simple sentences, indicating a dire situation.

———– EXTENDED ABSTRACT INTRODUCTION ———–Creole languages are low-resource languages, often genetically related to languages like English, French, and Portuguese, due to their linguistic histories with colonialism (DeGraff, 2003). As such, Creoles stand to benefit greatly from both data-efficient methods and transfer-learning from high-resource languages. At the same time, it has been observed by Lent et al. (2022b) that machine translation (MT) is a highly desired language technology by speakers of many Creoles. To this end, recent works have contributed new datasets, allowing for the development and evaluation of MT systems for Creoles (Robinson et al., 2024; Lent et al. 2024). In this work, we explore the use of the limited monolingual and parallel data for Creoles using parameter-efficient adaptation methods. Specifically, we compare the performance of different adapter architectures over the set of available benchmarks. We find adapters a promising approach for Creoles because they are parameter-efficient and have been shown to leverage transfer learning between related languages (Faisal and Anastasopoulos, 2022). While we perform experiments across multiple Creoles, we present only on Haitian Creole in this extended abstract. For future work, we aim to explore the potentials for leveraging other high-resourced languages for parameter-efficient transfer learning.

This study addresses the challenge of extending Large Language Models (LLMs) to non-English languages, specifically those using non-Roman scripts. We propose an approach that utilizes the romanized form of text as an interface for LLMs, hypothesizing that its frequent informal use and shared tokens with English enhance cross-lingual alignment. Our approach involve the continual pretraining of a English LLM like Llama 2 on romanized text of non-English, non-Roman script languages, followed by instruction tuning on romanized data. The results indicate that romanized text not only reduces token fertility by 2x-4x but also matches if not outperforms native script representation across various NLU, NLG and MT tasks. Moreover, the embeddings computed on romanized text exhibit closer alignment with their English translations than those from the native script. Our approach presents a promising direction for leveraging the power of English LLMs in languages traditionally underrepresented in NLP research.

Despite the considerable advancements in English LLMs, the progress in building comparable models for other languages has been hindered due to the scarcity of tailored resources. Our work aims to bridge this divide by introducing an expansive suite of resources specifically designed for the development of Indic LLMs, covering 22 languages, containing a total of 251B tokens and 74.8M instruction-response pairs. Recognizing the importance of both data quality and quantity, our approach combines highly curated manually verified data, unverified yet valuable data, and synthetic data. We build a clean, open-source pipeline for curating pre-training data from diverse sources, including websites, PDFs, and videos, incorporating best practices for crawling, cleaning, flagging, and deduplication. For instruction-fine tuning, we amalgamate existing Indic datasets, translate/transliterate English datasets into Indian languages, and utilize LLaMa2 and Mixtral models to create conversations grounded in articles from Indian Wikipedia and Wikihow. Additionally, we address toxicity alignment by generating toxic prompts for multiple scenarios and then generate non-toxic responses by feeding these toxic prompts to an aligned LLaMa2 model. We hope that the datasets, tools, and resources released as a part of this work will not only propel the research and development of Indic LLMs but also establish an open-source blueprint for extending such efforts to other languages.

While machine translation evaluation has been studied primarily for high-resource languages, there has been a recent interest in evaluation for low-resource languages due to the increasing availability of data and models. In this paper, we focus on a zero-shot evaluation setting focusing on low-resource Indian languages, namely Assamese, Kannada, Maithili, and Punjabi. We collect sufficient Multi-Dimensional Quality Metrics (MQM) and Direct Assessment (DA) annotations to create test sets and meta-evaluate a plethora of automatic evaluation metrics. We observe that even for learned metrics, which are known to exhibit zero-shot performance, the Kendall Tau and Pearson correlations with human annotations are only as high as 0.32 and 0.45. Synthetic data approaches show mixed results and overall do not help close the gap by much for these languages. This indicates that there is still a long way to go for low-resource evaluation.

The Nguni languages have over 20 million home language speakers in South Africa. There has been considerable growth in the datasets for Nguni languages, but so far no analysis of the performance of NLP models for these languages has been reported across languages and tasks. In this paper we study pretrained language models for the 4 Nguni languages - isiXhosa, isiZulu, isiNdebele, and Siswati. We compile publicly available datasets for natural language understanding and generation, spanning 6 tasks and 11 datasets. This benchmark, which we call NGLUEni, is the first centralised evaluation suite for the Nguni languages, allowing us to systematically evaluate the Nguni-language capabilities of pretrained language models (PLMs). Besides evaluating existing PLMs, we develop new PLMs for the Nguni languages through multilingual adaptive finetuning. Our models, Nguni-XLMR and Nguni-ByT5, outperform their base models and large-scale adapted models, showing that performance gains are obtainable through limited language group-based adaptation. We also perform experiments on cross-lingual transfer and machine translation. Our models achieve notable cross-lingual transfer improvements in the lower resourced Nguni languages (isiNdebele and Siswati). To facilitate future use of NGLUEni as a standardised evaluation suite for the Nguni languages, we create a web portal to access the collection of datasets and publicly release our models.

The rapid growth of machine translation (MT) systems necessitates meta-evaluations of evaluation metrics to enable selection of those that best reflect MT quality. Unfortunately, most meta-evaluation studies focus on European languages, the observations for which may not always apply to other languages. Indian languages, having over a billion speakers, are linguistically different from them, and to date, there are no such systematic studies focused solely on English to Indian language MT. This paper fills this gap through a Multidimensional Quality Metric (MQM) dataset consisting of 7000 fine-grained annotations, spanning 5 Indian languages and 7 MT systems. We evaluate 16 metrics and show that, pre-trained metrics like COMET have the highest correlations with annotator scores as opposed to n-gram metrics like BLEU. We further leverage our MQM annotations to develop an Indic-COMET metric and show that it outperforms COMET counterparts in both human scores correlations and robustness scores in Indian languages. Additionally, we show that the Indic-COMET can outperform COMET on some unseen Indian languages. We hope that our dataset and analysis will facilitate further research in Indic MT evaluation.

This paper studies the impact of layer normalization (LayerNorm) on zero-shot translation (ZST). Recent efforts for ZST often utilize the Transformer architecture as the backbone, with LayerNorm at the input of layers (PreNorm) set as the default. However, Xu et al. (2019) has revealed that PreNorm carries the risk of overfitting the training data. Based on this, we hypothesize that PreNorm may overfit supervised directions and thus have low generalizability for ZST. Through experiments on OPUS, IWSLT, and Europarl datasets for 54 ZST directions, we demonstrate that the original Transformer setting of LayerNorm after residual connections (PostNorm) consistently outperforms PreNorm by up to 12.3 BLEU points. We then study the performance disparities by analyzing the differences in off-target rates and structural variations between PreNorm and PostNorm. This study highlights the need for careful consideration of the LayerNorm setting for ZST.

In this paper, we present our open-source neural machine translation (NMT) toolkit called “Yet Another Neural Machine Translation Toolkit” abbreviated as YANMTT - https://github.com/prajdabre/yanmtt, which is built on top of the HuggingFace Transformers library. YANMTT focuses on transfer learning and enables easy pre-training and fine-tuning of sequence-to-sequence models at scale. It can be used for training parameter-heavy models with minimal parameter sharing and efficient, lightweight models via heavy parameter sharing. Additionally, it supports parameter-efficient fine-tuning (PEFT) through adapters and prompts. Our toolkit also comes with a user interface that can be used to demonstrate these models and visualize various parts of the model. Apart from these core features, our toolkit also provides other advanced functionalities such as but not limited to document/multi-source NMT, simultaneous NMT, mixtures-of-experts, model compression and continual learning.

There have been several meta-evaluation studies on the correlation between human ratings and offline machine translation (MT) evaluation metrics such as BLEU, chrF2, BertScore and COMET. These metrics have been used to evaluate simultaneous speech translation (SST) but their correlations with human ratings of SST, which has been recently collected as Continuous Ratings (CR), are unclear. In this paper, we leverage the evaluations of candidate systems submitted to the English-German SST task at IWSLT 2022 and conduct an extensive correlation analysis of CR and the aforementioned metrics. Our study reveals that the offline metrics are well correlated with CR and can be reliably used for evaluating machine translation in simultaneous mode, with some limitations on the test set size. We conclude that given the current quality levels of SST, these metrics can be used as proxies for CR, alleviating the need for large scale human evaluation. Additionally, we observe that correlations of the metrics with translation as a reference is significantly higher than with simultaneous interpreting, and thus we recommend the former for reliable evaluation.

Automatic speech translation is sensitive to speech recognition errors, but in a multilingual scenario, the same content may be available in various languages via simultaneous interpreting, dubbing or subtitling. In this paper, we hypothesize that leveraging multiple sources will improve translation quality if the sources complement one another in terms of correct information they contain. To this end, we first show that on a 10-hour ESIC corpus, the ASR errors in the original English speech and its simultaneous interpreting into German and Czech are mutually independent. We then use two sources, English and German, in a multi-source setting for translation into Czech to establish its robustness to ASR errors. Furthermore, we observe this robustness when translating both noisy sources together in a simultaneous translation setting. Our results show that multi-source neural machine translation has the potential to be useful in a real-time simultaneous translation setting, thereby motivating further investigation in this area.

Large language models have demonstrated the capability to perform on machine translation when the input is prompted with a few examples (in-context learning). Translation quality depends on various features of the selected examples, such as their quality and relevance, but previous work has predominantly focused on individual features in isolation. In this paper, we propose a general framework for combining different features influencing example selection. We learn a regression model, CTQ Scorer (Contextual Translation Quality), that selects examples based on multiple features in order to maximize the translation quality. On multiple language pairs and language models, we show that CTQ Scorer helps significantly outperform random selection as well as strong single-factor baselines reported in the literature. We also see an improvement of over 2.5 COMET points on average with respect to a strong BM25 retrieval-based baseline.

This study investigates machine translation between related languages i.e., languages within the same family that share linguistic characteristics such as word order and lexical similarity. Machine translation through few-shot prompting leverages a small set of translation pair examples to generate translations for test sentences. This procedure requires the model to learn how to generate translations while simultaneously ensuring that token ordering is maintained to produce a fluent and accurate translation. We propose that for related languages, the task of machine translation can be simplified by leveraging the monotonic alignment characteristic of such languages. We introduce DecoMT, a novel approach of few-shot prompting that decomposes the translation process into a sequence of word chunk translations. Through automatic and human evaluation conducted on multiple related language pairs across various language families, we demonstrate that our proposed approach of decomposed prompting surpasses multiple established few-shot baseline approaches. For example, DecoMT outperforms the strong few-shot prompting BLOOM model with an average improvement of 8 chrF++ scores across the examined languages.

This paper presents the results of the shared tasks from the 10th workshop on Asian translation (WAT2023). For the WAT2023, 2 teams submitted their translation results for the human evaluation. We also accepted 1 research paper. About 40 translation results were submitted to the automatic evaluation server, and selected submissions were manually evaluated.

Knowledge distillation (KD) is a well-known method for compressing neural models. However, works focusing on distilling knowledge from large multilingual neural machine translation (MNMT) models into smaller ones are practically nonexistent, despite the popularity and superiority of MNMT. This paper bridges this gap by presenting an empirical investigation of knowledge distillation for compressing MNMT models. We take Indic to English translation as a case study and demonstrate that commonly used language-agnostic and language-aware KD approaches yield models that are 4-5x smaller but also suffer from performance drops of up to 3.5 BLEU. To mitigate this, we then experiment with design considerations such as shallower versus deeper models, heavy parameter sharing, multistage training, and adapters. We observe that deeper compact models tend to be as good as shallower non-compact ones and that fine-tuning a distilled model on a high-quality subset slightly boosts translation quality. Overall, we conclude that compressing MNMT models via KD is challenging, indicating immense scope for further research.

In this paper, we (Team NICT-AI4B) describe our MT systems that we submit to the Indic MT task in WMT 2023. Our primary system consists of 3 stages: Joint denoising and MT training using officially approved monolingual and parallel corpora, backtranslation and, MT training on original and backtranslated parallel corpora. We observe that backtranslation leads to substantial improvements in translation quality up to 4 BLEU points. We also develop 2 contrastive systems on unconstrained settings, where the first system involves fine-tuning of IndicTrans2 DA models on official parallel corpora and seed data used in AI4Bharat et al, (2023), and the second system involves a system combination of the primary and the aforementioned system. Overall, we manage to obtain high-quality translation systems for the 4 low-resource North-East Indian languages of focus.

In this paper we evaluate the utility of large language models (LLMs) for translation of text with markup in which the most important and challenging aspect is to correctly transfer markup tags while ensuring that the content, both, inside and outside tags is correctly translated. While LLMs have been shown to be effective for plain text translation, their effectiveness for structured document translation is not well understood. To this end, we experiment with BLOOM and BLOOMZ, which are open-source multilingual LLMs, using zero, one and few-shot prompting, and compare with a domain-specific in-house NMT system using a detag-and-project approach for markup tags. We observe that LLMs with in-context learning exhibit poorer translation quality compared to the domain-specific NMT system, however, they are effective in transferring markup tags, especially the large BLOOM model (176 billion parameters). This is further confirmed by our human evaluation which also reveals the types of errors of the different tag transfer techniques. While LLM-based approaches come with the risk of losing, hallucinating and corrupting tags, they excel at placing them correctly in the translation.

This paper presents the results of the shared tasks from the 9th workshop on Asian translation (WAT2022). For the WAT2022, 8 teams submitted their translation results for the human evaluation. We also accepted 4 research papers. About 300 translation results were submitted to the automatic evaluation server, and selected submissions were manually evaluated.

We present our submission to the structured document translation task organized by WAT 2022. In structured document translation, the key challenge is the handling of inline tags, which annotate text. Specifically, the text that is annotated by tags, should be translated in such a way that in the translation should contain the tags annotating the translation. This challenge is further compounded by the lack of training data containing sentence pairs with inline XML tag annotated content. However, to our surprise, we find that existing multilingual NMT systems are able to handle the translation of text annotated with XML tags without any explicit training on data containing said tags. Specifically, massively multilingual translation models like M2M-100 perform well despite not being explicitly trained to handle structured content. This direct translation approach is often either as good as if not better than the traditional approach of “remove tag, translate and re-inject tag” also known as the “detag-and-project” approach.

Natural Language Generation (NLG) for non-English languages is hampered by the scarcity of datasets in these languages. We present the IndicNLG Benchmark, a collection of datasets for benchmarking NLG for 11 Indic languages. We focus on five diverse tasks, namely, biography generation using Wikipedia infoboxes, news headline generation, sentence summarization, paraphrase generation and, question generation. We describe the created datasets and use them to benchmark the performance of several monolingual and multilingual baselines that leverage pre-trained sequence-to-sequence models. Our results exhibit the strong performance of multilingual language-specific pre-trained models, and the utility of models trained on our dataset for other related NLG tasks. Our dataset creation methods can be easily applied to modest-resource languages as they involve simple steps such as scraping news articles and Wikipedia infoboxes, light cleaning, and pivoting through machine translation data. To the best of our knowledge, the IndicNLG Benchmark is the first NLG benchmark for Indic languages and the most diverse multilingual NLG dataset, with approximately 8M examples across 5 tasks and 11 languages. The datasets and models will be publicly available.

In this paper, we study pre-trained sequence-to-sequence models for a group of related languages, with a focus on Indic languages. We present IndicBART, a multilingual, sequence-to-sequence pre-trained model focusing on 11 Indic languages and English. IndicBART utilizes the orthographic similarity between Indic scripts to improve transfer learning between similar Indic languages. We evaluate IndicBART on two NLG tasks: Neural Machine Translation (NMT) and extreme summarization. Our experiments on NMT and extreme summarization show that a model specific to related languages like IndicBART is competitive with large pre-trained models like mBART50 despite being significantly smaller. It also performs well on very low-resource translation scenarios where languages are not included in pre-training or fine-tuning. Script sharing, multilingual training, and better utilization of limited model capacity contribute to the good performance of the compact IndicBART model.

Word alignment has proven to benefit many-to-many neural machine translation (NMT). However, high-quality ground-truth bilingual dictionaries were used for pre-editing in previous methods, which are unavailable for most language pairs. Meanwhile, the contrastive objective can implicitly utilize automatically learned word alignment, which has not been explored in many-to-many NMT. This work proposes a word-level contrastive objective to leverage word alignments for many-to-many NMT. Empirical results show that this leads to 0.8 BLEU gains for several language pairs. Analyses reveal that in many-to-many NMT, the encoder’s sentence retrieval performance highly correlates with the translation quality, which explains when the proposed method impacts translation. This motivates future exploration for many-to-many NMT to improve the encoder’s sentence retrieval performance.

In this paper, we describe KreolMorisienMT, a dataset for benchmarking machine translation quality of Mauritian Creole. Mauritian Creole (Kreol Morisien) is a French-based creole and a lingua franca of the Republic of Mauritius. KreolMorisienMT consists of a parallel corpus between English and Kreol Morisien, French and Kreol Morisien and a monolingual corpus for Kreol Morisien. We first give an overview of Kreol Morisien and then describe the steps taken to create the corpora. Thereafter, we benchmark Kreol Morisien ↔ English and Kreol Morisien ↔ French models leveraging pre-trained models and multilingual transfer learning. Human evaluation reveals our systems’ high translation quality.

Translation of structured content is an important application of machine translation, but the scarcity of evaluation data sets, especially for Asian languages, limits progress. In this paper we present a novel multilingual multiway evaluation data set for the translation of structured documents of the Asian languages Japanese, Korean and Chinese. We describe the data set, its creation process and important characteristics, followed by establishing and evaluating baselines using the direct translation as well as detag-project approaches. Our data set is well suited for multilingual evaluation, and it contains richer annotation tag sets than existing data sets. Our results show that massively multilingual translation models like M2M-100 and mBART-50 perform surprisingly well despite not being explicitly trained to handle structured content. The data set described in this paper and used in our experiments is released publicly.

Existing subword segmenters are either 1) frequency-based without semantics information or 2) neural-based but trained on parallel corpora. To address this, we present BERTSeg, an unsupervised neural subword segmenter for neural machine translation, which utilizes the contextualized semantic embeddings of words from characterBERT and maximizes the generation probability of subword segmentations. Furthermore, we propose a generation probability-based regularization method that enables BERTSeg to produce multiple segmentations for one word to improve the robustness of neural machine translation. Experimental results show that BERTSeg with regularization achieves up to 8 BLEU points improvement in 9 translation directions on ALT, IWSLT15 Vi->En, WMT16 Ro->En, and WMT15 Fi->En datasets compared with BPE. In addition, BERTSeg is efficient, needing up to 5 minutes for training.

In this paper, we describe our submission to the Code-mixed Machine Translation (MixMT) shared task. In MixMT, the objective is to translate Hinglish to English and vice versa. For our submissions, we focused on code-mixed pre-training and multi-way fine-tuning. Our submissions achieved rank 4 in terms of automatic evaluation score. For Hinglish to English translation, our submission achieved rank 4 as well.

In this paper we present FeatureBART, a linguistically motivated sequence-to-sequence monolingual pre-training strategy in which syntactic features such as lemma, part-of-speech and dependency labels are incorporated into the span prediction based pre-training framework (BART). These automatically extracted features are incorporated via approaches such as concatenation and relevance mechanisms, among which the latter is known to be better than the former. When used for low-resource NMT as a downstream task, we show that these feature based models give large improvements in bilingual settings and modest ones in multilingual settings over their counterparts that do not use features.

Neural machine translation (NMT) models are typically trained using a softmax cross-entropy loss where the softmax distribution is compared against the gold labels. In low-resource scenarios and NMT models tend to perform poorly because the model training quickly converges to a point where the softmax distribution computed using logits approaches the gold label distribution. Although label smoothing is a well-known solution to address this issue and we further propose to divide the logits by a temperature coefficient greater than one and forcing the softmax distribution to be smoother during training. This makes it harder for the model to quickly over-fit. In our experiments on 11 language pairs in the low-resource Asian Language Treebank dataset and we observed significant improvements in translation quality. Our analysis focuses on finding the right balance of label smoothing and softmax tempering which indicates that they are orthogonal methods. Finally and a study of softmax entropies and gradients reveal the impact of our method on the internal behavior of our NMT models.

In a real-time simultaneous translation setting and neural machine translation (NMT) models start generating target language tokens from incomplete source language sentences and making them harder to translate and leading to poor translation quality. Previous research has shown that document-level NMT and comprising of sentence and context encoders and a decoder and leverages context from neighboring sentences and helps improve translation quality. In simultaneous translation settings and the context from previous sentences should be even more critical. To this end and in this paper and we propose wait-k simultaneous document-level NMT where we keep the context encoder as it is and replace the source sentence encoder and target language decoder with their wait-k equivalents. We experiment with low and high resource settings using the ALT and OpenSubtitles2018 corpora and where we observe minor improvements in translation quality. We then perform an analysis of the translations obtained using our models by focusing on sentences that should benefit from the context where we found out that the model does and in fact and benefit from context but is unable to effectively leverage it and especially in a low-resource setting. This shows that there is a need for further innovation in the way useful context is identified and leveraged.

This paper presents the results of the shared tasks from the 8th workshop on Asian translation (WAT2021). For the WAT2021, 28 teams participated in the shared tasks and 24 teams submitted their translation results for the human evaluation. We also accepted 5 research papers. About 2,100 translation results were submitted to the automatic evaluation server, and selected submissions were manually evaluated.

In this paper we describe our submission to the multilingual Indic language translation wtask “MultiIndicMT” under the team name “NICT-5”. This task involves translation from 10 Indic languages into English and vice-versa. The objective of the task was to explore the utility of multilingual approaches using a variety of in-domain and out-of-domain parallel and monolingual corpora. Given the recent success of multilingual NMT pre-training we decided to explore pre-training an MBART model on a large monolingual corpus collection covering all languages in this task followed by multilingual fine-tuning on small in-domain corpora. Firstly, we observed that a small amount of pre-training followed by fine-tuning on small bilingual corpora can yield large gains over when pre-training is not used. Furthermore, multilingual fine-tuning leads to further gains in translation quality which significantly outperforms a very strong multilingual baseline that does not rely on any pre-training.

Cognates are variants of the same lexical form across different languages; for example “fonema” in Spanish and “phoneme” in English are cognates, both of which mean “a unit of sound”. The task of automatic detection of cognates among any two languages can help downstream NLP tasks such as Cross-lingual Information Retrieval, Computational Phylogenetics, and Machine Translation. In this paper, we demonstrate the use of cross-lingual word embeddings for detecting cognates among fourteen Indian Languages. Our approach introduces the use of context from a knowledge graph to generate improved feature representations for cognate detection. We, then, evaluate the impact of our cognate detection mechanism on neural machine translation (NMT), as a downstream task. We evaluate our methods to detect cognates on a challenging dataset of twelve Indian languages, namely, Sanskrit, Hindi, Assamese, Oriya, Kannada, Gujarati, Tamil, Telugu, Punjabi, Bengali, Marathi, and Malayalam. Additionally, we create evaluation datasets for two more Indian languages, Konkani and Nepali. We observe an improvement of up to 18% points, in terms of F-score, for cognate detection. Furthermore, we observe that cognates extracted using our method help improve NMT quality by up to 2.76 BLEU. We also release our code, newly constructed datasets and cross-lingual models publicly.

In this study, linguistic knowledge at different levels are incorporated into the neural machine translation (NMT) framework to improve translation quality for language pairs with extremely limited data. Integrating manually designed or automatically extracted features into the NMT framework is known to be beneficial. However, this study emphasizes that the relevance of the features is crucial to the performance. Specifically, we propose two methods, 1) self relevance and 2) word-based relevance, to improve the representation of features for NMT. Experiments are conducted on translation tasks from English to eight Asian languages, with no more than twenty thousand sentences for training. The proposed methods improve translation quality for all tasks by up to 3.09 BLEU points. Discussions with visualization provide the explainability of the proposed methods where we show that the relevance methods provide weights to features thereby enhancing their impact on low-resource machine translation.

The advent of neural machine translation (NMT) has opened up exciting research in building multilingual translation systems i.e. translation models that can handle more than one language pair. Many advances have been made which have enabled (1) improving translation for low-resource languages via transfer learning from high resource languages; and (2) building compact translation models spanning multiple languages. In this tutorial, we will cover the latest advances in NMT approaches that leverage multilingualism, especially to enhance low-resource translation. In particular, we will focus on the following topics: modeling parameter sharing for multi-way models, massively multilingual models, training protocols, language divergence, transfer learning, zero-shot/zero-resource learning, pivoting, multilingual pre-training and multi-source translation.

We propose a novel procedure for training multiple Transformers with tied parameters which compresses multiple models into one enabling the dynamic choice of the number of encoder and decoder layers during decoding. In training an encoder-decoder model, typically, the output of the last layer of the N-layer encoder is fed to the M-layer decoder, and the output of the last decoder layer is used to compute loss. Instead, our method computes a single loss consisting of NxM losses, where each loss is computed from the output of one of the M decoder layers connected to one of the N encoder layers. Such a model subsumes NxM models with different number of encoder and decoder layers, and can be used for decoding with fewer than the maximum number of encoder and decoder layers. Given our flexible tied model, we also address to a-priori selection of the number of encoder and decoder layers for faster decoding, and explore recurrent stacking of layers and knowledge distillation for model compression. We present a cost-benefit analysis of applying the proposed approaches for neural machine translation and show that they reduce decoding costs while preserving translation quality.

In neural machine translation (NMT), sequence distillation (SD) through creation of distilled corpora leads to efficient (compact and fast) models. However, its effectiveness in extremely low-resource (ELR) settings has not been well-studied. On the other hand, transfer learning (TL) by leveraging larger helping corpora greatly improves translation quality in general. This paper investigates a combination of SD and TL for training efficient NMT models for ELR settings, where we utilize TL with helping corpora twice: once for distilling the ELR corpora and then during compact model training. We experimented with two ELR settings: Vietnamese–English and Hindi–English from the Asian Language Treebank dataset with 18k training sentence pairs. Using the compact models with 40% smaller parameters trained on the distilled ELR corpora, greedy search achieved 3.6 BLEU points improvement in average while reducing 40% of decoding time. We also confirmed that using both the distilled ELR and helping corpora in the second round of TL further improves translation quality. Our work highlights the importance of stage-wise application of SD and TL for efficient NMT modeling for ELR settings.

Lectures translation is a case of spoken language translation and there is a lack of publicly available parallel corpora for this purpose. To address this, we examine a framework for parallel corpus mining which is a quick and effective way to mine a parallel corpus from publicly available lectures at Coursera. Our approach determines sentence alignments, relying on machine translation and cosine similarity over continuous-space sentence representations. We also show how to use the resulting corpora in a multistage fine-tuning based domain adaptation for high-quality lectures translation. For Japanese–English lectures translation, we extracted parallel data of approximately 40,000 lines and created development and test sets through manual filtering for benchmarking translation performance. We demonstrate that the mined corpus greatly enhances the quality of translation when used in conjunction with out-of-domain parallel corpora via multistage training. This paper also suggests some guidelines to gather and clean corpora, mine parallel sentences, address noise in the mined data, and create high-quality evaluation splits. For the sake of reproducibility, we have released our code for parallel data creation.

Neural machine translation (NMT) needs large parallel corpora for state-of-the-art translation quality. Low-resource NMT is typically addressed by transfer learning which leverages large monolingual or parallel corpora for pre-training. Monolingual pre-training approaches such as MASS (MAsked Sequence to Sequence) are extremely effective in boosting NMT quality for languages with small parallel corpora. However, they do not account for linguistic information obtained using syntactic analyzers which is known to be invaluable for several Natural Language Processing (NLP) tasks. To this end, we propose JASS, Japanese-specific Sequence to Sequence, as a novel pre-training alternative to MASS for NMT involving Japanese as the source or target language. JASS is joint BMASS (Bunsetsu MASS) and BRSS (Bunsetsu Reordering Sequence to Sequence) pre-training which focuses on Japanese linguistic units called bunsetsus. In our experiments on ASPEC Japanese–English and News Commentary Japanese–Russian translation we show that JASS can give results that are competitive with if not better than those given by MASS. Furthermore, we show for the first time that joint MASS and JASS pre-training gives results that significantly surpass the individual methods indicating their complementary nature. We will release our code, pre-trained models and bunsetsu annotated data as resources for researchers to use in their own NLP tasks.

Sequence-to-sequence (S2S) pre-training using large monolingual data is known to improve performance for various S2S NLP tasks. However, large monolingual corpora might not always be available for the languages of interest (LOI). Thus, we propose to exploit monolingual corpora of other languages to complement the scarcity of monolingual corpora for the LOI. We utilize script mapping (Chinese to Japanese) to increase the similarity (number of cognates) between the monolingual corpora of helping languages and LOI. An empirical case study of low-resource Japanese-English neural machine translation (NMT) reveals that leveraging large Chinese and French monolingual corpora can help overcome the shortage of Japanese and English monolingual corpora, respectively, for S2S pre-training. Using only Chinese and French monolingual corpora, we were able to improve Japanese-English translation quality by up to 8.5 BLEU in low-resource scenarios.

This paper presents the results of the shared tasks from the 7th workshop on Asian translation (WAT2020). For the WAT2020, 20 teams participated in the shared tasks and 14 teams submitted their translation results for the human evaluation. We also received 12 research paper submissions out of which 7 were accepted. About 500 translation results were submitted to the automatic evaluation server, and selected submissions were manually evaluated.

In this paper we describe our team‘s (NICT-5) Neural Machine Translation (NMT) models whose translations were submitted to shared tasks of the 7th Workshop on Asian Translation. We participated in the Indic language multilingual sub-task as well as the NICT-SAP multilingual multi-domain sub-task. We focused on naive many-to-many NMT models which gave reasonable translation quality despite their simplicity. Our observations are twofold: (a.) Many-to-many models suffer from a lack of consistency where the translation quality for some language pairs is very good but for some others it is terrible when compared against one-to-many and many-to-one baselines. (b.) Oversampling smaller corpora does not necessarily give the best translation quality for the language pair associated with that pair.

This paper highlights the impressive utility of multi-parallel corpora for transfer learning in a one-to-many low-resource neural machine translation (NMT) setting. We report on a systematic comparison of multistage fine-tuning configurations, consisting of (1) pre-training on an external large (209k–440k) parallel corpus for English and a helping target language, (2) mixed pre-training or fine-tuning on a mixture of the external and low-resource (18k) target parallel corpora, and (3) pure fine-tuning on the target parallel corpora. Our experiments confirm that multi-parallel corpora are extremely useful despite their scarcity and content-wise redundancy thus exhibiting the true power of multilingualism. Even when the helping target language is not one of the target languages of our concern, our multistage fine-tuning can give 3–9 BLEU score gains over a simple one-to-one model.

This paper presents the results of the shared tasks from the 6th workshop on Asian translation (WAT2019) including Ja↔En, Ja↔Zh scientific paper translation subtasks, Ja↔En, Ja↔Ko, Ja↔En patent translation subtasks, Hi↔En, My↔En, Km↔En, Ta↔En mixed domain subtasks and Ru↔Ja news commentary translation task. For the WAT2019, 25 teams participated in the shared tasks. We also received 10 research paper submissions out of which 61 were accepted. About 400 translation results were submitted to the automatic evaluation server, and selected submis- sions were manually evaluated.

In this paper we describe our submissions to WAT 2019 for the following tasks: English–Tamil translation and Russian–Japanese translation. Our team,“NICT-5”, focused on multilingual domain adaptation and back-translation for Russian–Japanese translation and on simple fine-tuning for English–Tamil translation . We noted that multi-stage fine tuning is essential in leveraging the power of multilingualism for an extremely low-resource language like Russian–Japanese. Furthermore, we can improve the performance of such a low-resource language pair by exploiting a small but in-domain monolingual corpus via back-translation. We managed to obtain second rank in both tasks for all translation directions.

In this paper, we describe our supervised neural machine translation (NMT) systems that we developed for the news translation task for Kazakh↔English, Gujarati↔English, Chinese↔English, and English→Finnish translation directions. We focused on leveraging multilingual transfer learning and back-translation for the extremely low-resource language pairs: Kazakh↔English and Gujarati↔English translation. For the Chinese↔English translation, we used the provided parallel data augmented with a large quantity of back-translated monolingual data to train state-of-the-art NMT systems. We then employed techniques that have been proven to be most effective, such as back-translation, fine-tuning, and model ensembling, to generate the primary submissions of Chinese↔English. For English→Finnish, our submission from WMT18 remains a strong baseline despite the increase in parallel corpora for this year’s task.

In this paper we describe our neural machine translation (NMT) systems for Japanese↔English translation which we submitted to the translation robustness task. We focused on leveraging transfer learning via fine tuning to improve translation quality. We used a fairly well established domain adaptation technique called Mixed Fine Tuning (MFT) (Chu et. al., 2017) to improve translation quality for Japanese↔English. We also trained bi-directional NMT models instead of uni-directional ones as the former are known to be quite robust, especially in low-resource scenarios. However, given the noisy nature of the in-domain training data, the improvements we obtained are rather modest.

This paper presents the NICT’s participation in the WMT19 shared Similar Language Translation Task. We participated in the Spanish-Portuguese task. For both translation directions, we prepared state-of-the-art statistical (SMT) and neural (NMT) machine translation systems. Our NMT systems with the Transformer architecture were trained on the provided parallel data enlarged with a large quantity of back-translated monolingual data. Our primary submission to the task is the result of a simple combination of our SMT and NMT systems. According to BLEU, our systems were ranked second and third respectively for the Portuguese-to-Spanish and Spanish-to-Portuguese translation directions. For contrastive experiments, we also submitted outputs generated with an unsupervised SMT system.

Machine Translation (MT) is a sub-field of NLP which has experienced a number of paradigm shifts since its inception. Up until 2014, Phrase Based Statistical Machine Translation (PBSMT) approaches used to be the state of the art. In late 2014, Neural Machine Translation (NMT) was introduced and was proven to outperform all PBSMT approaches by a significant margin. Since then, the NMT approaches have undergone several transformations which have pushed the state of the art even further. This tutorial is primarily aimed at researchers who are either interested in or are fairly new to the world of NMT and want to obtain a deep understanding of NMT fundamentals. Because it will also cover the latest developments in NMT, it should also be useful to attendees with some experience in NMT.

We describe here our Machine Translation (MT) model and the results we obtained for the IWSLT 2017 Multilingual Shared Task. Motivated by Zero Shot NMT [1] we trained a Multilingual Neural Machine Translation by combining all the training data into one single collection by appending the tokens to the source sentences in order to indicate the target language they should be translated to. We observed that even in a low resource situation we were able to get translations whose quality surpass the quality of those obtained by Phrase Based Statistical Machine Translation by several BLEU points. The most surprising result we obtained was in the zero shot setting for Dutch-German and Italian-Romanian where we observed that despite using no parallel corpora between these language pairs, the NMT model was able to translate between these languages and the translations were either as good as or better (in terms of BLEU) than the non zero resource setting. We also verify that the NMT models that use feed forward layers and self attention instead of recurrent layers are extremely fast in terms of training which is useful in a NMT experimental setting.

We describe here our approaches and results on the WAT 2017 shared translation tasks. Following our good results with Neural Machine Translation in the previous shared task, we continue this approach this year, with incremental improvements in models and training methods. We focused on the ASPEC dataset and could improve the state-of-the-art results for Chinese-to-Japanese and Japanese-to-Chinese translations.

In this paper, we propose a novel domain adaptation method named “mixed fine tuning” for neural machine translation (NMT). We combine two existing approaches namely fine tuning and multi domain NMT. We first train an NMT model on an out-of-domain parallel corpus, and then fine tune it on a parallel corpus which is a mix of the in-domain and out-of-domain corpora. All corpora are augmented with artificial tags to indicate specific domains. We empirically compare our proposed method against fine tuning and multi domain methods and discuss its benefits and shortcomings.

Parallel corpora are crucial for machine translation (MT), however they are quite scarce for most language pairs and domains. As comparable corpora are far more available, many studies have been conducted to extract parallel sentences from them for MT. In this paper, we exploit the neural network features acquired from neural MT for parallel sentence extraction. We observe significant improvements for both accuracy in sentence extraction and MT performance.

India is a country with 22 officially recognized languages and 17 of these have WordNets, a crucial resource. Web browser based interfaces are available for these WordNets, but are not suited for mobile devices which deters people from effectively using this resource. We present our initial work on developing mobile applications and browser extensions to access WordNets for Indian Languages. Our contribution is two fold: (1) We develop mobile applications for the Android, iOS and Windows Phone OS platforms for Hindi, Marathi and Sanskrit WordNets which allow users to search for words and obtain more information along with their translations in English and other Indian languages. (2) We also develop browser extensions for English, Hindi, Marathi, and Sanskrit WordNets, for both Mozilla Firefox, and Google Chrome. We believe that such applications can be quite helpful in a classroom scenario, where students would be able to access the WordNets as dictionaries as well as lexical knowledge bases. This can help in overcoming the language barrier along with furthering language understanding.