Jan Niehues

Sai Koneru, Miriam Exel, Matthias Huck et al.

Large Language Models (LLM's) have demonstrated considerable success in various Natural Language Processing tasks, but they have yet to attain state-of-the-art performance in Neural Machine Translation (NMT). Nevertheless, their significant performance in tasks demanding a broad understanding and contextual processing shows their potential for translation. To exploit these abilities, we investigate using LLM's for MT and explore recent parameter-efficient fine-tuning techniques. Surprisingly, our initial experiments find that fine-tuning for translation purposes even led to performance degradation. To overcome this, we propose an alternative approach: adapting LLM's as Automatic Post-Editors (APE) rather than direct translators. Building on the LLM's exceptional ability to process and generate lengthy sequences, we also propose extending our approach to document-level translation. We show that leveraging Low-Rank-Adapter fine-tuning for APE can yield significant improvements across both sentence and document-level metrics while generalizing to out-of-domain data. Most notably, we achieve a state-of-the-art accuracy rate of 89\% on the ContraPro test set, which specifically assesses the model's ability to resolve pronoun ambiguities when translating from English to German. Lastly, we investigate a practical scenario involving manual post-editing for document-level translation, where reference context is made available. Here, we demonstrate that leveraging human corrections can significantly reduce the number of edits required for subsequent translations (Interactive Demo for integrating manual feedback can be found here: https://huggingface.co/spaces/skoneru/contextual_refinement_ende).

Peter Polák, Ngoc-Quan Ngoc, Tuan-Nam Nguyen et al.

In this paper, we describe our submission to the Simultaneous Speech Translation at IWSLT 2022. We explore strategies to utilize an offline model in a simultaneous setting without the need to modify the original model. In our experiments, we show that our onlinization algorithm is almost on par with the offline setting while being $3\times$ faster than offline in terms of latency on the test set. We also show that the onlinized offline model outperforms the best IWSLT2021 simultaneous system in medium and high latency regimes and is almost on par in the low latency regime. We make our system publicly available.

Ngoc-Quan Pham, Alex Waibel, Jan Niehues

Multilingual speech recognition with supervised learning has achieved great results as reflected in recent research. With the development of pretraining methods on audio and text data, it is imperative to transfer the knowledge from unsupervised multilingual models to facilitate recognition, especially in many languages with limited data. Our work investigated the effectiveness of using two pretrained models for two modalities: wav2vec 2.0 for audio and MBART50 for text, together with the adaptive weight techniques to massively improve the recognition quality on the public datasets containing CommonVoice and Europarl. Overall, we noticed an 44% improvement over purely supervised learning, and more importantly, each technique provides a different reinforcement in different languages. We also explore other possibilities to potentially obtain the best model by slightly adding either depth or relative attention to the architecture.

Christian Huber, Tu Anh Dinh, Carlos Mullov et al.

The challenge of low-latency speech translation has recently draw significant interest in the research community as shown by several publications and shared tasks. Therefore, it is essential to evaluate these different approaches in realistic scenarios. However, currently only specific aspects of the systems are evaluated and often it is not possible to compare different approaches. In this work, we propose the first framework to perform and evaluate the various aspects of low-latency speech translation under realistic conditions. The evaluation is carried out in an end-to-end fashion. This includes the segmentation of the audio as well as the run-time of the different components. Secondly, we compare different approaches to low-latency speech translation using this framework. We evaluate models with the option to revise the output as well as methods with fixed output. Furthermore, we directly compare state-of-the-art cascaded as well as end-to-end systems. Finally, the framework allows to automatically evaluate the translation quality as well as latency and also provides a web interface to show the low-latency model outputs to the user.

Danni Liu, Thai Binh Nguyen, Sai Koneru et al.

Many existing speech translation benchmarks focus on native-English speech in high-quality recording conditions, which often do not match the conditions in real-life use-cases. In this paper, we describe our speech translation system for the multilingual track of IWSLT 2023, which evaluates translation quality on scientific conference talks. The test condition features accented input speech and terminology-dense contents. The task requires translation into 10 languages of varying amounts of resources. In absence of training data from the target domain, we use a retrieval-based approach (kNN-MT) for effective adaptation (+0.8 BLEU for speech translation). We also use adapters to easily integrate incremental training data from data augmentation, and show that it matches the performance of re-training. We observe that cascaded systems are more easily adaptable towards specific target domains, due to their separate modules. Our cascaded speech system substantially outperforms its end-to-end counterpart on scientific talk translation, although their performance remains similar on TED talks.

Ngoc-Quan Pham, Jan Niehues, Alexander Waibel

Multilingual speech recognition with neural networks is often implemented with batch-learning, when all of the languages are available before training. An ability to add new languages after the prior training sessions can be economically beneficial, but the main challenge is catastrophic forgetting. In this work, we combine the qualities of weight factorization and elastic weight consolidation in order to counter catastrophic forgetting and facilitate learning new languages quickly. Such combination allowed us to eliminate catastrophic forgetting while still achieving performance for the new languages comparable with having all languages at once, in experiments of learning from an initial 10 languages to achieve 26 languages without catastrophic forgetting and a reasonable performance compared to training all languages from scratch.

Siqi Li, Danni Liu, Jan Niehues

Direct speech translation (ST) models often struggle with rare words. Incorrect translation of these words can have severe consequences, impacting translation quality and user trust. While rare word translation is inherently challenging for neural models due to sparse learning signals, real-world scenarios often allow access to translations of past recordings on similar topics. To leverage these valuable resources, we propose a retrieval-and-demonstration approach to enhance rare word translation accuracy in direct ST models. First, we adapt existing ST models to incorporate retrieved examples for rare word translation, which allows the model to benefit from prepended examples, similar to in-context learning. We then develop a cross-modal (speech-to-speech, speech-to-text, text-to-text) retriever to locate suitable examples. We demonstrate that standard ST models can be effectively adapted to leverage examples for rare word translation, improving rare word translation accuracy over the baseline by 17.6% with gold examples and 8.5% with retrieved examples. Moreover, our speech-to-speech retrieval approach outperforms other modalities and exhibits higher robustness to unseen speakers. Our code is publicly available (https://github.com/SiqiLii/Retrieve-and-Demonstration-ST).

Danni Liu, Jan Niehues

Customizing machine translation models to comply with desired attributes (e.g., formality or grammatical gender) is a well-studied topic. However, most current approaches rely on (semi-)supervised data with attribute annotations. This data scarcity bottlenecks democratizing such customization possibilities to a wider range of languages, particularly lower-resource ones. This gap is out of sync with recent progress in pretrained massively multilingual translation models. In response, we transfer the attribute controlling capabilities to languages without attribute-annotated data with an NLLB-200 model as a foundation. Inspired by techniques from controllable generation, we employ a gradient-based inference-time controller to steer the pretrained model. The controller transfers well to zero-shot conditions, as it operates on pretrained multilingual representations and is attribute -- rather than language-specific. With a comprehensive comparison to finetuning-based control, we demonstrate that, despite finetuning's clear dominance in supervised settings, the gap to inference-time control closes when moving to zero-shot conditions, especially with new and distant target languages. The latter also shows stronger domain robustness. We further show that our inference-time control complements finetuning. A human evaluation on a real low-resource language, Bengali, confirms our findings. Our code is https://github.com/dannigt/attribute-controller-transfer

Danni Liu, Jan Niehues

The cornerstone of multilingual neural translation is shared representations across languages. Given the theoretically infinite representation power of neural networks, semantically identical sentences are likely represented differently. While representing sentences in the continuous latent space ensures expressiveness, it introduces the risk of capturing of irrelevant features which hinders the learning of a common representation. In this work, we discretize the encoder output latent space of multilingual models by assigning encoder states to entries in a codebook, which in effect represents source sentences in a new artificial language. This discretization process not only offers a new way to interpret the otherwise black-box model representations, but, more importantly, gives potential for increasing robustness in unseen testing conditions. We validate our approach on large-scale experiments with realistic data volumes and domains. When tested in zero-shot conditions, our approach is competitive with two strong alternatives from the literature. We also use the learned artificial language to analyze model behavior, and discover that using a similar bridge language increases knowledge-sharing among the remaining languages.

Pedro Jeuris, Jan Niehues

Recently, we have seen an increasing interest in the area of speech-to-text translation. This has led to astonishing improvements in this area. In contrast, the activities in the area of speech-to-speech translation is still limited, although it is essential to overcome the language barrier. We believe that one of the limiting factors is the availability of appropriate training data. We address this issue by creating LibriS2S, to our knowledge the first publicly available speech-to-speech training corpus between German and English. For this corpus, we used independently created audio for German and English leading to an unbiased pronunciation of the text in both languages. This allows the creation of a new text-to-speech and speech-to-speech translation model that directly learns to generate the speech signal based on the pronunciation of the source language. Using this created corpus, we propose Text-to-Speech models based on the example of the recently proposed FastSpeech 2 model that integrates source language information. We do this by adapting the model to take information such as the pitch, energy or transcript from the source speech as additional input.

Shashank Subramanya, Jan Niehues

Applications designed for simultaneous speech translation during events such as conferences or meetings need to balance quality and lag while displaying translated text to deliver a good user experience. One common approach to building online spoken language translation systems is by leveraging models built for offline speech translation. Based on a technique to adapt end-to-end monolingual models, we investigate multilingual models and different architectures (end-to-end and cascade) on the ability to perform online speech translation. On the multilingual TEDx corpus, we show that the approach generalizes to different architectures. We see similar gains in latency reduction (40% relative) across languages and architectures. However, the end-to-end architecture leads to smaller translation quality losses after adapting to the online model. Furthermore, the approach even scales to zero-shot directions.

Zhaolin Li, Jan Niehues

When building state-of-the-art speech translation models, the need for large computational resources is a significant obstacle due to the large training data size and complex models. The availability of pre-trained models is a promising opportunity to build strong speech translation systems efficiently. In a first step, we investigate efficient strategies to build cascaded and end-to-end speech translation systems based on pre-trained models. Using this strategy, we can train and apply the models on a single GPU. While the end-to-end models show superior translation performance to cascaded ones, the application of this technology has a limitation on the need for additional end-to-end training data. In a second step, we proposed an additional similarity loss to encourage the model to generate similar hidden representations for speech and transcript. Using this technique, we can increase the data efficiency and improve the translation quality by 6 BLEU points in scenarios with limited end-to-end training data.

Sai Koneru, Fabian Retkowski, Christian Huber et al.

The globalization of education and rapid growth of online learning have made localizing educational content a critical challenge. Lecture materials are inherently multimodal, combining spoken audio with visual slides, which requires systems capable of processing multiple input modalities. To provide an accessible and complete learning experience, translations must preserve all modalities: text for reading, slides for visual understanding, and speech for auditory learning. We present \textbf{BOOM}, a multimodal multilingual lecture companion that jointly translates lecture audio and slides to produce synchronized outputs across three modalities: translated text, localized slides with preserved visual elements, and synthesized speech. This end-to-end approach enables students to access lectures in their native language while aiming to preserve the original content in its entirety. Our experiments demonstrate that slide-aware transcripts also yield cascading benefits for downstream tasks such as summarization and question answering. We release our Slide Translation code at https://github.com/saikoneru/image-translator and integrate it in Lecture Translator at https://gitlab.kit.edu/kit/isl-ai4lt/lt-middleware/ltpipeline}\footnote{All released code and models are licensed under the MIT License.

Supriti Sinhamahapatra, Thai-Binh Nguyen, Yiğit Oğuz et al.

The goal of multilingual speech technology is to facilitate seamless communication between individuals speaking different languages, creating the experience as though everyone were a multilingual speaker. To create this experience, speech technology needs to address several challenges: Handling mixed multilingual input, specific vocabulary, and code-switching. However, there is currently no dataset benchmarking this situation. We propose a new benchmark to evaluate current Automatic Speech Recognition (ASR) systems, whether they are able to handle these challenges. The benchmark consists of bilingual discussions on scientific papers between multiple speakers, each conversing in a different language. We provide a standard evaluation framework, beyond Word Error Rate (WER) enabling consistent comparison of ASR performance across languages. Experimental results demonstrate that the proposed dataset is still an open challenge for state-of-the-art ASR systems. The dataset is available in https://huggingface.co/datasets/goodpiku/muscat-eval \\ \newline \Keywords{multilingual, speech recognition, audio segmentation, speaker diarization}

Sai Koneru, Matthias Huck, Miriam Exel et al.

Recent advancements in NLP have resulted in models with specialized strengths, such as processing multimodal inputs or excelling in specific domains. However, real-world tasks, like multimodal translation, often require a combination of these strengths, such as handling both translation and image processing. While individual translation and vision models are powerful, they typically lack the ability to perform both tasks in a single system. Combining these models poses challenges, particularly due to differences in their vocabularies, which limit the effectiveness of traditional ensemble methods to post-generation techniques like N-best list re-ranking. In this work, we propose a novel zero-shot ensembling strategy that allows for the integration of different models during the decoding phase without the need for additional training. Our approach re-ranks beams during decoding by combining scores at the word level, using heuristics to predict when a word is completed. We demonstrate the effectiveness of this method in machine translation scenarios, showing that it enables the generation of translations that are both speech- and image-aware while also improving overall translation quality (We will release the code upon paper acceptance.).

Robin Amann, Zhaolin Li, Barbara Bruno et al.

Speech disfluency commonly occurs in conversational and spontaneous speech. However, standard Automatic Speech Recognition (ASR) models struggle to accurately recognize these disfluencies because they are typically trained on fluent transcripts. Current research mainly focuses on detecting disfluencies within transcripts, overlooking their exact location and duration in the speech. Additionally, previous work often requires model fine-tuning and addresses limited types of disfluencies. In this work, we present an inference-only approach to augment any ASR model with the ability to detect open-set disfluencies. We first demonstrate that ASR models have difficulty transcribing speech disfluencies. Next, this work proposes a modified Connectionist Temporal Classification(CTC)-based forced alignment algorithm from \cite{kurzinger2020ctc} to predict word-level timestamps while effectively capturing disfluent speech. Additionally, we develop a model to classify alignment gaps between timestamps as either containing disfluent speech or silence. This model achieves an accuracy of 81.62% and an F1-score of 80.07%. We test the augmentation pipeline of alignment gap detection and classification on a disfluent dataset. Our results show that we captured 74.13% of the words that were initially missed by the transcription, demonstrating the potential of this pipeline for downstream tasks.

Renhan Lou, Jan Niehues

In machine translation, a common problem is that the translation of certain words even if translated can cause incomprehension of the target language audience due to different cultural backgrounds. A solution to solve this problem is to add explanations for these words. In a first step, we therefore need to identify these words or phrases. In this work we explore techniques to extract example explanations from a parallel corpus. However, the sparsity of sentences containing words that need to be explained makes building the training dataset extremely difficult. In this work, we propose a semi-automatic technique to extract these explanations from a large parallel corpus. Experiments on English->German language pair show that our method is able to extract sentence so that more than 10% of the sentences contain explanation, while only 1.9% of the original sentences contain explanations. In addition, experiments on English->French and English->Chinese language pairs also show similar conclusions. This is therefore an essential first automatic step to create a explanation dataset. Furthermore we show that the technique is robust for all three language pairs.

Vladimir Solovyev, Danni Liu, Jan Niehues

Finetuning pretrained models on downstream generation tasks often leads to catastrophic forgetting in zero-shot conditions. In this work, we focus on summarization and tackle the problem through the lens of language-independent representations. After training on monolingual summarization, we perform zero-shot transfer to new languages or language pairs. We first show naively finetuned models are highly language-specific in both output behavior and internal representations, resulting in poor zero-shot performance. Next, we propose query-key (QK) finetuning to decouple task-specific knowledge from the pretrained language generation abilities. Then, after showing downsides of the standard adversarial language classifier, we propose a balanced variant that more directly enforces language-agnostic representations. Moreover, our qualitative analyses show removing source language identity correlates to zero-shot summarization performance. Our code is openly available.

Ibrahim Said Ahmad, Antonios Anastasopoulos, Ondřej Bojar et al.

This paper reports on the shared tasks organized by the 21st IWSLT Conference. The shared tasks address 7 scientific challenges in spoken language translation: simultaneous and offline translation, automatic subtitling and dubbing, speech-to-speech translation, dialect and low-resource speech translation, and Indic languages. The shared tasks attracted 18 teams whose submissions are documented in 26 system papers. The growing interest towards spoken language translation is also witnessed by the constantly increasing number of shared task organizers and contributors to the overview paper, almost evenly distributed across industry and academia.

Gabriel R. Rosenbaum, Lavender Yao Jiang, Ivaxi Sheth et al.

Large language models (LLMs) have recently emerged as powerful tools, finding many medical applications. LLMs' ability to coalesce vast amounts of information from many sources to generate a response-a process similar to that of a human expert-has led many to see potential in deploying LLMs for clinical use. However, medicine is a setting where accurate reasoning is paramount. Many researchers are questioning the effectiveness of multiple choice question answering (MCQA) benchmarks, frequently used to test LLMs. Researchers and clinicians alike must have complete confidence in LLMs' abilities for them to be deployed in a medical setting. To address this need for understanding, we introduce a knowledge graph (KG)-based method to evaluate the biomedical reasoning abilities of LLMs. Essentially, we map how LLMs link medical concepts in order to better understand how they reason. We test GPT-4, Llama3-70b, and PalmyraMed-70b, a specialized medical model. We enlist a panel of medical students to review a total of 60 LLM-generated graphs and compare these graphs to BIOS, a large biomedical KG. We observe GPT-4 to perform best in our human review but worst in our ground truth comparison; vice-versa with PalmyraMed, the medical model. Our work provides a means of visualizing the medical reasoning pathways of LLMs so they can be implemented in clinical settings safely and effectively.

Isik Baran Sandan, Tu Anh Dinh, Jan Niehues

Large Language Models (LLMs) have shown to be effective evaluators across various domains such as machine translations or the scientific domain. Current LLM-as-a-Judge approaches rely mostly on individual assessments or a single round of pairwise assessments, preventing the judge LLM from developing a global ranking perspective. To address this, we present Knockout Assessment, an LLM-asa Judge method using a knockout tournament system with iterative pairwise comparisons. Experiments across three LLMs on two datasets show that knockout assessment improves scoring accuracy, increasing Pearson correlation with expert evaluations by 0.07 on average for university-level exam scoring and machine translation evaluations, aligning LLM assessments more closely with human scoring.

Fabian Retkowski, Maike Züfle, Andreas Sudmann et al.

Speech summarization has become an essential tool for efficiently managing and accessing the growing volume of spoken and audiovisual content. However, despite its increasing importance, speech summarization remains loosely defined. The field intersects with several research areas, including speech recognition, text summarization, and specific applications like meeting summarization. This survey not only examines existing datasets and evaluation protocols, which are crucial for assessing the quality of summarization approaches, but also synthesizes recent developments in the field, highlighting the shift from traditional systems to advanced models like fine-tuned cascaded architectures and end-to-end solutions. In doing so, we surface the ongoing challenges, such as the need for realistic evaluation benchmarks, multilingual datasets, and long-context handling.

Maike Züfle, Sara Papi, Beatrice Savoldi et al.

Scientific communication is receiving increasing attention in natural language processing, especially to help researches access, summarize, and generate content. One emerging application in this area is Speech-to-Abstract Generation (SAG), which aims to automatically generate abstracts from recorded scientific presentations. SAG enables researchers to efficiently engage with conference talks, but progress has been limited by a lack of large-scale datasets. To address this gap, we introduce NUTSHELL, a novel multimodal dataset of *ACL conference talks paired with their corresponding abstracts. We establish strong baselines for SAG and evaluate the quality of generated abstracts using both automatic metrics and human judgments. Our results highlight the challenges of SAG and demonstrate the benefits of training on NUTSHELL. By releasing NUTSHELL under an open license (CC-BY 4.0), we aim to advance research in SAG and foster the development of improved models and evaluation methods.

Tu Anh Dinh, Tobias Palzer, Jan Niehues

Providing quality scores along with Machine Translation (MT) output, so-called reference-free Quality Estimation (QE), is crucial to inform users about the reliability of the translation. We propose a model-specific, unsupervised QE approach, termed $k$NN-QE, that extracts information from the MT model's training data using $k$-nearest neighbors. Measuring the performance of model-specific QE is not straightforward, since they provide quality scores on their own MT output, thus cannot be evaluated using benchmark QE test sets containing human quality scores on premade MT output. Therefore, we propose an automatic evaluation method that uses quality scores from reference-based metrics as gold standard instead of human-generated ones. We are the first to conduct detailed analyses and conclude that this automatic method is sufficient, and the reference-based MetricX-23 is best for the task.

Sai Koneru, Maike Züfle, Thai-Binh Nguyen et al.

The scope of the International Workshop on Spoken Language Translation (IWSLT) has recently broadened beyond traditional Speech Translation (ST) to encompass a wider array of tasks, including Speech Question Answering and Summarization. This shift is partly driven by the growing capabilities of modern systems, particularly with the success of Large Language Models (LLMs). In this paper, we present the Karlsruhe Institute of Technology's submissions for the Offline ST and Instruction Following (IF) tracks, where we leverage LLMs to enhance performance across all tasks. For the Offline ST track, we propose a pipeline that employs multiple automatic speech recognition systems, whose outputs are fused using an LLM with document-level context. This is followed by a two-step translation process, incorporating additional refinement step to improve translation quality. For the IF track, we develop an end-to-end model that integrates a speech encoder with an LLM to perform a wide range of instruction-following tasks. We complement it with a final document-level refinement stage to further enhance output quality by using contextual information.

Sai Koneru, Matthias Huck, Miriam Exel et al.

Quality Estimation (QE) models for Neural Machine Translation (NMT) predict the quality of the hypothesis without having access to the reference. An emerging research direction in NMT involves the use of QE models, which have demonstrated high correlations with human judgment and can enhance translations through Quality-Aware Decoding. Although several approaches have been proposed based on sampling multiple candidate translations and picking the best candidate, none have integrated these models directly into the decoding process. In this paper, we address this by proposing a novel token-level QE model capable of reliably scoring partial translations. We build a uni-directional QE model for this, as decoder models are inherently trained and efficient on partial sequences. We then present a decoding strategy that integrates the QE model for Quality-Aware decoding and demonstrate that the translation quality improves when compared to the N-best list re-ranking with state-of-the-art QE models (up to $1.39$ XCOMET-XXL $\uparrow$). Finally, we show that our approach provides significant benefits in document translation tasks, where the quality of N-best lists is typically suboptimal. Code can be found at https://ai4lt.iar.kit.edu/english/projects\_kontextmt.php

Danni Liu, Jan Niehues

Fine-tuning multilingual foundation models on specific languages often induces catastrophic forgetting, degrading performance on languages unseen in fine-tuning. While this phenomenon is widely-documented, the literature presents fragmented results about when forgetting occurs. To address this ambiguity, we conduct a systematic empirical study using machine translation as a testbed to identify the conditions that trigger catastrophic forgetting in multilingual fine-tuning. Through controlled experiments across different model architectures, data scales, and fine-tuning approaches, we reveal that the relative scale between model and data size is a primary determinant of forgetting. Moreover, we demonstrate that a model's instruction-following ability is more critical for retaining multilingual knowledge than its architecture. Contrary to assumptions, parameter-efficient fine-tuning offers no clear advantage over full fine-tuning in mitigating forgetting. Lastly, we show that cross-lingual alignment can mitigate forgetting while also facilitating positive transfer to unseen target languages.

Tu Anh Dinh, Jan Niehues

Quality Estimation (QE) is estimating quality of the model output during inference when the ground truth is not available. Deriving output quality from the models' output probability is the most trivial and low-effort way. However, we show that the output probability of text-generation models can appear underconfident. At each output step, there can be multiple correct options, making the probability distribution spread out more. Thus, lower probability does not necessarily mean lower output quality. Due to this observation, we propose a QE approach called BoostedProb, which boosts the model's confidence in cases where there are multiple viable output options. With no increase in complexity, BoostedProb is notably better than raw model probability in different settings, achieving on average +0.194 improvement in Pearson correlation to ground-truth quality. It also comes close to or outperforms more costly approaches like supervised or ensemble-based QE in certain settings.

Sai Koneru, Thai-Binh Nguyen, Ngoc-Quan Pham et al.

Large Language Models (LLMs) are currently under exploration for various tasks, including Automatic Speech Recognition (ASR), Machine Translation (MT), and even End-to-End Speech Translation (ST). In this paper, we present KIT's offline submission in the constrained + LLM track by incorporating recently proposed techniques that can be added to any cascaded speech translation. Specifically, we integrate Mistral-7B\footnote{mistralai/Mistral-7B-Instruct-v0.1} into our system to enhance it in two ways. Firstly, we refine the ASR outputs by utilizing the N-best lists generated by our system and fine-tuning the LLM to predict the transcript accurately. Secondly, we refine the MT outputs at the document level by fine-tuning the LLM, leveraging both ASR and MT predictions to improve translation quality. We find that integrating the LLM into the ASR and MT systems results in an absolute improvement of $0.3\%$ in Word Error Rate and $0.65\%$ in COMET for tst2019 test set. In challenging test sets with overlapping speakers and background noise, we find that integrating LLM is not beneficial due to poor ASR performance. Here, we use ASR with chunked long-form decoding to improve context usage that may be unavailable when transcribing with Voice Activity Detection segmentation alone.

Tu Anh Dinh, Carlos Mullov, Leonard Bärmann et al.

With the rapid development of Large Language Models (LLMs), it is crucial to have benchmarks which can evaluate the ability of LLMs on different domains. One common use of LLMs is performing tasks on scientific topics, such as writing algorithms, querying databases or giving mathematical proofs. Inspired by the way university students are evaluated on such tasks, in this paper, we propose SciEx - a benchmark consisting of university computer science exam questions, to evaluate LLMs ability on solving scientific tasks. SciEx is (1) multilingual, containing both English and German exams, and (2) multi-modal, containing questions that involve images, and (3) contains various types of freeform questions with different difficulty levels, due to the nature of university exams. We evaluate the performance of various state-of-the-art LLMs on our new benchmark. Since SciEx questions are freeform, it is not straightforward to evaluate LLM performance. Therefore, we provide human expert grading of the LLM outputs on SciEx. We show that the free-form exams in SciEx remain challenging for the current LLMs, where the best LLM only achieves 59.4\% exam grade on average. We also provide detailed comparisons between LLM performance and student performance on SciEx. To enable future evaluation of new LLMs, we propose using LLM-as-a-judge to grade the LLM answers on SciEx. Our experiments show that, although they do not perform perfectly on solving the exams, LLMs are decent as graders, achieving 0.948 Pearson correlation with expert grading.

Matthias Sperber, Ondřej Bojar, Barry Haddow et al.

Human evaluation is a critical component in machine translation system development and has received much attention in text translation research. However, little prior work exists on the topic of human evaluation for speech translation, which adds additional challenges such as noisy data and segmentation mismatches. We take first steps to fill this gap by conducting a comprehensive human evaluation of the results of several shared tasks from the last International Workshop on Spoken Language Translation (IWSLT 2023). We propose an effective evaluation strategy based on automatic resegmentation and direct assessment with segment context. Our analysis revealed that: 1) the proposed evaluation strategy is robust and scores well-correlated with other types of human judgements; 2) automatic metrics are usually, but not always, well-correlated with direct assessment scores; and 3) COMET as a slightly stronger automatic metric than chrF, despite the segmentation noise introduced by the resegmentation step systems. We release the collected human-annotated data in order to encourage further investigation.

Lena Cabrera, Jan Niehues

Neural machine translation (NMT) models often suffer from gender biases that harm users and society at large. In this work, we explore how bridging the gap between languages for which parallel data is not available affects gender bias in multilingual NMT, specifically for zero-shot directions. We evaluate translation between grammatical gender languages which requires preserving the inherent gender information from the source in the target language. We study the effect of encouraging language-agnostic hidden representations on models' ability to preserve gender and compare pivot-based and zero-shot translation regarding the influence of the bridge language (participating in all language pairs during training) on gender preservation. We find that language-agnostic representations mitigate zero-shot models' masculine bias, and with increased levels of gender inflection in the bridge language, pivoting surpasses zero-shot translation regarding fairer gender preservation for speaker-related gender agreement.

Tu Anh Dinh, Jan Niehues

Quality Estimation (QE) is the task of predicting the quality of Machine Translation (MT) system output, without using any gold-standard translation references. State-of-the-art QE models are supervised: they require human-labeled quality of some MT system output on some datasets for training, making them domain-dependent and MT-system-dependent. There has been research on unsupervised QE, which requires glass-box access to the MT systems, or parallel MT data to generate synthetic errors for training QE models. In this paper, we present Perturbation-based QE - a word-level Quality Estimation approach that works simply by analyzing MT system output on perturbed input source sentences. Our approach is unsupervised, explainable, and can evaluate any type of blackbox MT systems, including the currently prominent large language models (LLMs) with opaque internal processes. For language directions with no labeled QE data, our approach has similar or better performance than the zero-shot supervised approach on the WMT21 shared task. Our approach is better at detecting gender bias and word-sense-disambiguation errors in translation than supervised QE, indicating its robustness to out-of-domain usage. The performance gap is larger when detecting errors on a nontraditional translation-prompting LLM, indicating that our approach is more generalizable to different MT systems. We give examples demonstrating our approach's explainability power, where it shows which input source words have influence on a certain MT output word.

Zhong Zhou, Jan Niehues, Alex Waibel

In many humanitarian scenarios, translation into severely low resource languages often does not require a universal translation engine, but a dedicated text-specific translation engine. For example, healthcare records, hygienic procedures, government communication, emergency procedures and religious texts are all limited texts. While generic translation engines for all languages do not exist, translation of multilingually known limited texts into new, endangered languages may be possible and reduce human translation effort. We attempt to leverage translation resources from many rich resource languages to efficiently produce best possible translation quality for a well known text, which is available in multiple languages, in a new, severely low resource language. We examine two approaches: 1. best selection of seed sentences to jump start translations in a new language in view of best generalization to the remainder of a larger targeted text(s), and 2. we adapt large general multilingual translation engines from many other languages to focus on a specific text in a new, unknown language. We find that adapting large pretrained multilingual models to the domain/text first and then to the severely low resource language works best. If we also select a best set of seed sentences, we can improve average chrF performance on new test languages from a baseline of 21.9 to 50.7, while reducing the number of seed sentences to only around 1,000 in the new, unknown language.

Tu Anh Dinh, Danni Liu, Jan Niehues

Recently, end-to-end speech translation (ST) has gained significant attention as it avoids error propagation. However, the approach suffers from data scarcity. It heavily depends on direct ST data and is less efficient in making use of speech transcription and text translation data, which is often more easily available. In the related field of multilingual text translation, several techniques have been proposed for zero-shot translation. A main idea is to increase the similarity of semantically similar sentences in different languages. We investigate whether these ideas can be applied to speech translation, by building ST models trained on speech transcription and text translation data. We investigate the effects of data augmentation and auxiliary loss function. The techniques were successfully applied to few-shot ST using limited ST data, with improvements of up to +12.9 BLEU points compared to direct end-to-end ST and +3.1 BLEU points compared to ST models fine-tuned from ASR model.

Sai Koneru, Danni Liu, Jan Niehues

While Active Learning (AL) techniques are explored in Neural Machine Translation (NMT), only a few works focus on tackling low annotation budgets where a limited number of sentences can get translated. Such situations are especially challenging and can occur for endangered languages with few human annotators or having cost constraints to label large amounts of data. Although AL is shown to be helpful with large budgets, it is not enough to build high-quality translation systems in these low-resource conditions. In this work, we propose a cost-effective training procedure to increase the performance of NMT models utilizing a small number of annotated sentences and dictionary entries. Our method leverages monolingual data with self-supervised objectives and a small-scale, inexpensive dictionary for additional supervision to initialize the NMT model before applying AL. We show that improving the model using a combination of these knowledge sources is essential to exploit AL strategies and increase gains in low-resource conditions. We also present a novel AL strategy inspired by domain adaptation for NMT and show that it is effective for low budgets. We propose a new hybrid data-driven approach, which samples sentences that are diverse from the labelled data and also most similar to unlabelled data. Finally, we show that initializing the NMT model and further using our AL strategy can achieve gains of up to $13$ BLEU compared to conventional AL methods.

Sai Koneru, Danni Liu, Jan Niehues

Unsupervised neural machine translation (UNMT) is beneficial especially for low resource languages such as those from the Dravidian family. However, UNMT systems tend to fail in realistic scenarios involving actual low resource languages. Recent works propose to utilize auxiliary parallel data and have achieved state-of-the-art results. In this work, we focus on unsupervised translation between English and Kannada, a low resource Dravidian language. We additionally utilize a limited amount of auxiliary data between English and other related Dravidian languages. We show that unifying the writing systems is essential in unsupervised translation between the Dravidian languages. We explore several model architectures that use the auxiliary data in order to maximize knowledge sharing and enable UNMT for distant language pairs. Our experiments demonstrate that it is crucial to include auxiliary languages that are similar to our focal language, Kannada. Furthermore, we propose a metric to measure language similarity and show that it serves as a good indicator for selecting the auxiliary languages.

Jan Niehues

While recent advances in deep learning led to significant improvements in machine translation, neural machine translation is often still not able to continuously adapt to the environment. For humans, as well as for machine translation, bilingual dictionaries are a promising knowledge source to continuously integrate new knowledge. However, their exploitation poses several challenges: The system needs to be able to perform one-shot learning as well as model the morphology of source and target language. In this work, we proposed an evaluation framework to assess the ability of neural machine translation to continuously learn new phrases. We integrate one-shot learning methods for neural machine translation with different word representations and show that it is important to address both in order to successfully make use of bilingual dictionaries. By addressing both challenges we are able to improve the ability to translate new, rare words and phrases from 30% to up to 70%. The correct lemma is even generated by more than 90%.

Danni Liu, Jan Niehues, James Cross et al.

Multilingual neural machine translation has shown the capability of directly translating between language pairs unseen in training, i.e. zero-shot translation. Despite being conceptually attractive, it often suffers from low output quality. The difficulty of generalizing to new translation directions suggests the model representations are highly specific to those language pairs seen in training. We demonstrate that a main factor causing the language-specific representations is the positional correspondence to input tokens. We show that this can be easily alleviated by removing residual connections in an encoder layer. With this modification, we gain up to 18.5 BLEU points on zero-shot translation while retaining quality on supervised directions. The improvements are particularly prominent between related languages, where our proposed model outperforms pivot-based translation. Moreover, our approach allows easy integration of new languages, which substantially expands translation coverage. By thorough inspections of the hidden layer outputs, we show that our approach indeed leads to more language-independent representations.

Danni Liu, Jan Niehues, Gerasimos Spanakis

Automatic speech recognition (ASR) systems are primarily evaluated on transcription accuracy. However, in some use cases such as subtitling, verbatim transcription would reduce output readability given limited screen size and reading time. Therefore, this work focuses on ASR with output compression, a task challenging for supervised approaches due to the scarcity of training data. We first investigate a cascaded system, where an unsupervised compression model is used to post-edit the transcribed speech. We then compare several methods of end-to-end speech recognition under output length constraints. The experiments show that with limited data far less than needed for training a model from scratch, we can adapt a Transformer-based ASR model to incorporate both transcription and compression capabilities. Furthermore, the best performance in terms of WER and ROUGE scores is achieved by explicitly modeling the length constraints within the end-to-end ASR system.

Danni Liu, Gerasimos Spanakis, Jan Niehues

Encoder-decoder models provide a generic architecture for sequence-to-sequence tasks such as speech recognition and translation. While offline systems are often evaluated on quality metrics like word error rates (WER) and BLEU, latency is also a crucial factor in many practical use-cases. We propose three latency reduction techniques for chunk-based incremental inference and evaluate their efficiency in terms of accuracy-latency trade-off. On the 300-hour How2 dataset, we reduce latency by 83% to 0.8 second by sacrificing 1% WER (6% rel.) compared to offline transcription. Although our experiments use the Transformer, the hypothesis selection strategies are applicable to other encoder-decoder models. To avoid expensive re-computation, we use a unidirectionally-attending encoder. After an adaptation procedure to partial sequences, the unidirectional model performs on-par with the original model. We further show that our approach is also applicable to low-latency speech translation. On How2 English-Portuguese speech translation, we reduce latency to 0.7 second (-84% rel.) while incurring a loss of 2.4 BLEU points (5% rel.) compared to the offline system.

Ngoc-Quan Pham, Thanh-Le Ha, Tuan-Nam Nguyen et al.

Transformer models are powerful sequence-to-sequence architectures that are capable of directly mapping speech inputs to transcriptions or translations. However, the mechanism for modeling positions in this model was tailored for text modeling, and thus is less ideal for acoustic inputs. In this work, we adapt the relative position encoding scheme to the Speech Transformer, where the key addition is relative distance between input states in the self-attention network. As a result, the network can better adapt to the variable distributions present in speech data. Our experiments show that our resulting model achieves the best recognition result on the Switchboard benchmark in the non-augmentation condition, and the best published result in the MuST-C speech translation benchmark. We also show that this model is able to better utilize synthetic data than the Transformer, and adapts better to variable sentence segmentation quality for speech translation.

Jan Niehues

We have seen significant improvements in machine translation due to the usage of deep learning. While the improvements in translation quality are impressive, the encoder-decoder architecture enables many more possibilities. In this paper, we explore one of these, the generation of constraint translation. We focus on length constraints, which are essential if the translation should be displayed in a given format. In this work, we propose an end-to-end approach for this task. Compared to a traditional method that first translates and then performs sentence compression, the text compression is learned completely unsupervised. By combining the idea with zero-shot multilingual machine translation, we are also able to perform unsupervised monolingual sentence compression. In order to fulfill the length constraints, we investigated several methods to integrate the constraints into the model. Using the presented technique, we are able to significantly improve the translation quality under constraints. Furthermore, we are able to perform unsupervised monolingual sentence compression.

Thai Son Nguyen, Jan Niehues, Eunah Cho et al.

User studies have shown that reducing the latency of our simultaneous lecture translation system should be the most important goal. We therefore have worked on several techniques for reducing the latency for both components, the automatic speech recognition and the speech translation module. Since the commonly used commitment latency is not appropriate in our case of continuous stream decoding, we focused on word latency. We used it to analyze the performance of our current system and to identify opportunities for improvements. In order to minimize the latency we combined run-on decoding with a technique for identifying stable partial hypotheses when stream decoding and a protocol for dynamic output update that allows to revise the most recent parts of the transcription. This combination reduces the latency at word level, where the words are final and will never be updated again in the future, from 18.1s to 1.1s without sacrificing performance in terms of word error rate.

Thai-Son Nguyen, Sebastian Stueker, Jan Niehues et al.

Sequence-to-Sequence (S2S) models recently started to show state-of-the-art performance for automatic speech recognition (ASR). With these large and deep models overfitting remains the largest problem, outweighing performance improvements that can be obtained from better architectures. One solution to the overfitting problem is increasing the amount of available training data and the variety exhibited by the training data with the help of data augmentation. In this paper we examine the influence of three data augmentation methods on the performance of two S2S model architectures. One of the data augmentation method comes from literature, while two other methods are our own development - a time perturbation in the frequency domain and sub-sequence sampling. Our experiments on Switchboard and Fisher data show state-of-the-art performance for S2S models that are trained solely on the speech training data and do not use additional text data.

Jan Niehues, Ngoc-Quan Pham

Recently, significant improvements have been achieved in various natural language processing tasks using neural sequence-to-sequence models. While aiming for the best generation quality is important, ultimately it is also necessary to develop models that can assess the quality of their output. In this work, we propose to use the similarity between training and test conditions as a measure for models' confidence. We investigate methods solely using the similarity as well as methods combining it with the posterior probability. While traditionally only target tokens are annotated with confidence measures, we also investigate methods to annotate source tokens with confidence. By learning an internal alignment model, we can significantly improve confidence projection over using state-of-the-art external alignment tools. We evaluate the proposed methods on downstream confidence estimation for machine translation (MT). We show improvements on segment-level confidence estimation as well as on confidence estimation for source tokens. In addition, we show that the same methods can also be applied to other tasks using sequence-to-sequence models. On the automatic speech recognition (ASR) task, we are able to find 60% of the errors by looking at 20% of the data.

Stefan Constantin, Jan Niehues, Alex Waibel

The state-of-the-art neural network architectures make it possible to create spoken language understanding systems with high quality and fast processing time. One major challenge for real-world applications is the high latency of these systems caused by triggered actions with high executions times. If an action can be separated into subactions, the reaction time of the systems can be improved through incremental processing of the user utterance and starting subactions while the utterance is still being uttered. In this work, we present a model-agnostic method to achieve high quality in processing incrementally produced partial utterances. Based on clean and noisy versions of the ATIS dataset, we show how to create datasets with our method to create low-latency natural language understanding components. We get improvements of up to 47.91 absolute percentage points in the metric F1-score.

Ngoc-Quan Pham, Jan Niehues, Thanh-Le Ha et al.

An important concern in training multilingual neural machine translation (NMT) is to translate between language pairs unseen during training, i.e zero-shot translation. Improving this ability kills two birds with one stone by providing an alternative to pivot translation which also allows us to better understand how the model captures information between languages. In this work, we carried out an investigation on this capability of the multilingual NMT models. First, we intentionally create an encoder architecture which is independent with respect to the source language. Such experiments shed light on the ability of NMT encoders to learn multilingual representations, in general. Based on such proof of concept, we were able to design regularization methods into the standard Transformer model, so that the whole architecture becomes more robust in zero-shot conditions. We investigated the behaviour of such models on the standard IWSLT 2017 multilingual dataset. We achieved an average improvement of 2.23 BLEU points across 12 language pairs compared to the zero-shot performance of a state-of-the-art multilingual system. Additionally, we carry out further experiments in which the effect is confirmed even for language pairs with multiple intermediate pivots.

Ngoc-Quan Pham, Thai-Son Nguyen, Jan Niehues et al.

Recently, end-to-end sequence-to-sequence models for speech recognition have gained significant interest in the research community. While previous architecture choices revolve around time-delay neural networks (TDNN) and long short-term memory (LSTM) recurrent neural networks, we propose to use self-attention via the Transformer architecture as an alternative. Our analysis shows that deep Transformer networks with high learning capacity are able to exceed performance from previous end-to-end approaches and even match the conventional hybrid systems. Moreover, we trained very deep models with up to 48 Transformer layers for both encoder and decoders combined with stochastic residual connections, which greatly improve generalizability and training efficiency. The resulting models outperform all previous end-to-end ASR approaches on the Switchboard benchmark. An ensemble of these models achieve 9.9% and 17.7% WER on Switchboard and CallHome test sets respectively. This finding brings our end-to-end models to competitive levels with previous hybrid systems. Further, with model ensembling the Transformers can outperform certain hybrid systems, which are more complicated in terms of both structure and training procedure.

Matthias Sperber, Graham Neubig, Jan Niehues et al.

Speech translation has traditionally been approached through cascaded models consisting of a speech recognizer trained on a corpus of transcribed speech, and a machine translation system trained on parallel texts. Several recent works have shown the feasibility of collapsing the cascade into a single, direct model that can be trained in an end-to-end fashion on a corpus of translated speech. However, experiments are inconclusive on whether the cascade or the direct model is stronger, and have only been conducted under the unrealistic assumption that both are trained on equal amounts of data, ignoring other available speech recognition and machine translation corpora. In this paper, we demonstrate that direct speech translation models require more data to perform well than cascaded models, and while they allow including auxiliary data through multi-task training, they are poor at exploiting such data, putting them at a severe disadvantage. As a remedy, we propose the use of end-to-end trainable models with two attention mechanisms, the first establishing source speech to source text alignments, the second modeling source to target text alignment. We show that such models naturally decompose into multi-task-trainable recognition and translation tasks and propose an attention-passing technique that alleviates error propagation issues in a previous formulation of a model with two attention stages. Our proposed model outperforms all examined baselines and is able to exploit auxiliary training data much more effectively than direct attentional models.