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.

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.

With the number of scientific papers published every year growing and current large language models (LLMs) showing state-of-the-art performance on natural language processing (NLP) tasks, we ask the question if LLMs could be utilized to answer questions on scientific papers.We investigate how well state-of-the-art large language models (LLMs) can answer questions on scientific paper by experimenting with long-context versions of the LLaMA 2 model and evaluating and training on the Qasper dataset.We analyze how well the LLMs handle longer papers and questions that can only be answered by accessing information from far out paragraphs. During our experiments, we see that the performance of these LLMs drops with growing length and position of relevant information.We employ different measures from simple prompts to chain-of-thought prompts and zero-shot usage to fine-tuning with QLoRA.While we still observe a performance loss with increased context length, our measures reduce the effects of this flaw, and we can achieve F₁ scores similar to bigger models like GPT-4.

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.

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 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.

This paper presents KIT’s submissions to the IWSLT 2024 dialectal and low-resource track. In this work, we build systems for translating into English from speech in Maltese, Bemba, and two Arabic dialects Tunisian and North Levantine. Under the unconstrained condition, we leverage the pre-trained multilingual models by fine-tuning them for the target language pairs to address data scarcity problems in this track. We build cascaded and end-to-end speech translation systems for different language pairs and show the cascaded system brings slightly better overall performance. Besides, we find utilizing additional data resources boosts speech recognition performance but slightly harms machine translation performance in cascaded systems. Lastly, we show that Minimum Bayes Risk is effective in improving speech translation performance by combining the cascaded and end-to-end systems, bringing a consistent improvement of around 1 BLUE point.

Speech translation has witnessed significant progress driven by advancements in modeling techniques and the growing availability of training data. In this paper, we highlight recent advances in two ongoing research directions in ST: scaling the models to 1) many translation directions (multilingual ST) and 2) beyond the text output modality (multimodal ST). We structure this review by examining the sequential stages of a model’s development lifecycle: determining training resources, selecting model architecture, training procedures, evaluation metrics, and deployment considerations. We aim to highlight recent developments in each stage, with a particular focus on model architectures (dedicated speech translation models and LLM-based general-purpose model) and training procedures (task-specific vs. task-invariant approaches). Based on the reviewed advancements, we identify and discuss ongoing challenges within the field of speech translation.

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 is 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. Moreover, a human evaluation on a real low-resource language, Bengali, confirms our findings. Our code is in the supplementary material.

Diffusion models have recently shown great potential on many generative tasks.In this work, we explore diffusion models for machine translation (MT).We adapt two prominent diffusion-based text generation models, Diffusion-LM and DiffuSeq, to perform machine translation.As the diffusion models generate non-autoregressively (NAR),we draw parallels to NAR machine translation models.With a comparison to conventional Transformer-based translation models, as well as to the Levenshtein Transformer,an established NAR MT model,we show that the multimodality problem that limits NAR machine translation performance is also a challenge to diffusion models.We demonstrate that knowledge distillation from an autoregressive model improves the performance of diffusion-based MT.A thorough analysis on the translation quality of inputs of different lengths shows that the diffusion models struggle more on long-range dependencies than other models.

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 focuses on the translation of scientific conference talks. The test condition features accented input speech and terminology-dense contents. The tasks 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 outperforms its end-to-end counterpart on scientific talk translation, although their performance remains similar on TED talks.

In this paper, we describe our submission to the Simultaneous Track at IWSLT 2023. This year, we continue with the successful setup from the last year, however, we adopt the latest methods that further improve the translation quality. Additionally, we propose a novel online policy for attentional encoder-decoder models. The policy prevents the model to generate translation beyond the current speech input by using an auxiliary CTC output layer. We show that the proposed simultaneous policy can be applied to both streaming blockwise models and offline encoder-decoder models. We observe significant improvements in quality (up to 1.1 BLEU) and the computational footprint (up to 45% relative RTF).

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.

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.

Pretrained models in acoustic and textual modalities can potentially improve speech translation for both Cascade and End-to-end approaches. In this evaluation, we aim at empirically looking for the answer by using the wav2vec, mBART50 and DeltaLM models to improve text and speech translation models. The experiments showed that the presence of these models together with an advanced audio segmentation method results in an improvement over the previous end-to-end system by up to 7 BLEU points. More importantly, the experiments showed that given enough data and modeling capacity to overcome the training difficulty, we can outperform even very competitive Cascade systems. In our experiments, this gap can be as large as 2.0 BLEU points, the same gap that the Cascade often led over the years.

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 3x 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.

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.

This paper describes Maastricht University’s participation in the IWSLT 2021 multilingual speech translation track. The task in this track is to build multilingual speech translation systems in supervised and zero-shot directions. Our primary system is an end-to-end model that performs both speech transcription and translation. We observe that the joint training for the two tasks is complementary especially when the speech translation data is scarce. On the source and target side, we use data augmentation and pseudo-labels respectively to improve the performance of our systems. We also introduce an ensembling technique that consistently improves the quality of transcriptions and translations. The experiments show that the end-to-end system is competitive with its cascaded counterpart especially in zero-shot conditions.

We present our development of the multilingual machine translation system for the large-scale multilingual machine translation task at WMT 2021. Starting form the provided baseline system, we investigated several techniques to improve the translation quality on the target subset of languages. We were able to significantly improve the translation quality by adapting the system towards the target subset of languages and by generating synthetic data using the initial model. Techniques successfully applied in zero-shot multilingual machine translation (e.g. similarity regularizer) only had a minor effect on the final translation performance.

Unsupervised Neural Machine translation (UNMT) is beneficial especially for under-resourced languages such as from the Dravidian family. They learn to translate between the source and target, relying solely on only monolingual corpora. However, UNMT systems fail in scenarios that occur often when dealing with low resource languages. Recent works have achieved state-of-the-art results by adding auxiliary parallel data with similar languages. In this work, we focus on unsupervised translation between English and Kannada by using limited amounts of auxiliary data between English and other Dravidian languages. We show that transliteration is essential in unsupervised translation between Dravidian languages, as they do not share a common writing system. We explore several model architectures that use the auxiliary data in order to maximize knowledge sharing and enable UNMT for dissimilar language pairs. We show from our experiments it is crucial for Kannada and reference languages to be similar. Further, we propose a method to measure language similarity to choose the most beneficial reference languages.

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.