The increasing use of ML in astronomy introduces important questions about interpretability. Due to their complexity and non-linear nature, it can be challenging to understand their decision-making process. While these models can effectively identify unusual spectra, interpreting the physical nature of the flagged outliers remains a major challenge. We aim to bridge the gap between anomaly detection and physical understanding by combining deep learning with interpretable ML (iML) techniques to identify and explain anomalous galaxy spectra from SDSS data. We present a flexible framework that uses a variational autoencoder to compute multiple anomaly scores, including physically-motivated variants of the mean squared error. We adapt the iML LIME algorithm to spectroscopic data, systematically explore segmentation and perturbation strategies, and compute explanation weights that identify the features most responsible for each detection. To uncover population-level trends, we normalize the LIME weights and apply clustering to the top 1\% most anomalous spectra. Our approach successfully separates instrumental artifacts from physically meaningful outliers and groups anomalous spectra into astrophysically coherent categories. These include dusty, metal-rich starbursts; chemically-enriched H\,II regions with moderate excitation; and extreme emission-line galaxies with low metallicity and hard ionizing spectra. The explanation weights align with established emission-line diagnostics, enabling a physically-grounded taxonomy of spectroscopic anomalies. Our work shows that interpretable anomaly detection provides a scalable, transparent, and physically meaningful approach to exploring large spectroscopic datasets. Our framework opens the door for incorporating interpretability tools into quality control, follow-up targeting, and discovery pipelines in current and future surveys.
Understanding the decision-making processes of neural networks is a central goal of mechanistic interpretability. In the context of Large Language Models (LLMs), this involves uncovering the underlying mechanisms and identifying the roles of individual model components such as neurons and attention heads, as well as model abstractions such as the learned sparse features extracted by Sparse Autoencoders (SAEs). A rapidly growing line of work tackles this challenge by using powerful generator models to produce open-vocabulary, natural language concept descriptions for these components. In this paper, we provide the first survey of the emerging field of concept descriptions for model components and abstractions. We chart the key methods for generating these descriptions, the evolving landscape of automated and human metrics for evaluating them, and the datasets that underpin this research. Our synthesis reveals a growing demand for more rigorous, causal evaluation. By outlining the state of the art and identifying key challenges, this survey provides a roadmap for future research toward making models more transparent.
Machine Interpreting systems are currently implemented as unimodal, real-time speech-to-speech architectures, processing translation exclusively on the basis of the linguistic signal. Such reliance on a single modality, however, constrains performance in contexts where disambiguation and adequacy depend on additional cues, such as visual, situational, or pragmatic information. This paper introduces Vision-Grounded Interpreting (VGI), a novel approach designed to address the limitations of unimodal machine interpreting. We present a prototype system that integrates a vision-language model to process both speech and visual input from a webcam, with the aim of priming the translation process through contextual visual information. To evaluate the effectiveness of this approach, we constructed a hand-crafted diagnostic corpus targeting three types of ambiguity. In our evaluation, visual grounding substantially improves lexical disambiguation, yields modest and less stable gains for gender resolution, and shows no benefit for syntactic ambiguities. We argue that embracing multimodality represents a necessary step forward for advancing translation quality in machine interpreting.
Rust aims to offer full memory safety for programs, a guarantee that untamed C programs do not enjoy. How difficult is it to translate existing C code to Rust? To get a complementary view from that of automatic C to Rust translators, we report on a user study asking humans to translate real-world C programs to Rust. Our participants are able to produce safe Rust translations, whereas state-of-the-art automatic tools are not able to do so. Our analysis highlights that the high-level strategy taken by users departs significantly from those of automatic tools we study. We also find that users often choose zero-cost (static) abstractions for temporal safety, which addresses a predominant component of runtime costs in other full memory safety defenses. User-provided translations showcase a rich landscape of specialized strategies to translate the same C program in different ways to safe Rust, which future automatic translators can consider.
In the field of node representation learning the task of interpreting latent dimensions has become a prominent, well-studied research topic. The contribution of this work focuses on appraising the interpretability of another rarely-exploited feature of node embeddings increasingly utilised in recommendation and consumption diversity studies: inter-node embedded distances. Introducing a new method to measure how understandable the distances between nodes are, our work assesses how well the proximity weights derived from a network before embedding relate to the node closeness measurements after embedding. Testing several classical node embedding models, our findings reach a conclusion familiar to practitioners albeit rarely cited in literature - the matrix factorisation model SVD is the most interpretable through 1, 2 and even higher-order proximities.
This study aims to provide insights into understanding the theoretical background of the application of critical discourse analysis (CDA) in the translation of political texts in the field of translation studies. The study also casts light on the investigation into the ideological and discursive issues in translation through the use of CDA as well as political discourse and translation. CDA is crucial in understanding the role and significance of discourse in the translation of a political text without disregarding the literary sense, authentic style of the speaker in the target language, and rhetorical devices. In this regard, this study considers the case of a political speech to demonstrate the role and significance of CDA in the translation of political speech. For this reason, the study has selected the case of Donald Trump’s inaugural address for translation into the target language of Turkish by the study’s author through the use of a critical lens. Following a critical approach and Norman Fairclough’s (1995) model for CDA in the interpretation and translation of political discourse, this study aims to provide explanations and solutions to the difficulties encountered in the interpretation and translation of a political speech. Therefore, the comparison of the source text with the target text offered and discussed in this study helps to underline and raise awareness about the contributions of CDA to translation studies.
This issue of Revista Tradumàtica explores how technology, including machine translation, AI, and accessibility tools, transforms professional translation. Articles address psychological impacts, productivity, quality, and usability. Highlights include autonomy’s link to job satisfaction, stress from concurrent workflows, and challenges with large language models and remote interpreting platforms. Accessibility studies emphasize user involvement in design. While technology boosts productivity, it introduces stress and uncertainty, underscoring the importance of user-driven development to enhance satisfaction, autonomy, and translation quality.
This paper explores the job satisfaction of translators working for an international intergovernmental organization. The extant literature on translators’ job satisfaction has explored a number of constructs. Based on developments in the field of organizational theory and the complexity of translation as a job, it is argued that psychological ownership may prove an adequate framework to explain translators’ job satisfaction and instrumental in establishing a dialogue between the various analyses of different workplaces in the field of translation and interpreting studies. The study focuses on a specific multilingual intergovernmental organization and draws on the interviews of 17 Spanish-native translators of different nationalities. Their feelings of ownership are analyzed and variations in how they relate to constructs of psychological ownership — feelings of control, intimacy of knowledge, and self-investment— become apparent. Exploring patterns shows those variations to be related to translators’ differing translation dosas, that is, they're divergent, competing, and sometimes conflicting understanding of what translation is and should be. Furthermore, relationships between psychological ownership, translation doxa, and translators’ efforts to advance their own doxas in the organization are examined with a view towards creating means to engage professional translators in advancing a doxa shaped by and for translators across workplaces.
El sector del vino posee un léxico propio, no solo en lo que a terminología especializada se refiere, sino también en lo que respecta a los referentes culturales que aparecen en los textos que se producen en este ámbito. En España, los vinos se producen en diferentes regiones con características culturales particulares, que utilizan procesos distintos a los de otros países. Estos culturemas se utilizan a menudo en los textos comerciales sobre vino como estrategias comerciales para crear una determinada imagen sobre España, como se mostrará a través del análisis de dos sitios web en inglés, que permitirá observar las técnicas de traducción empleadas para trasvasar los culturemas y acercar la cultura vitivinícola española al público anglófono y determinar, así, las estrategias más adecuadas de traducción para estas referencias culturales.
Miguel Rios, Raluca-Maria Chereji, Alina Secara
et al.
Multilingual Neural Machine Translation (MNMT) models leverage many language pairs during training to improve translation quality for low-resource languages by transferring knowledge from high-resource languages. We study the quality of a domain-adapted MNMT model in the medical domain for English-Romanian with automatic metrics and a human error typology annotation which includes terminology-specific error categories. We compare the out-of-domain MNMT with the in-domain adapted MNMT. The in-domain MNMT model outperforms the out-of-domain MNMT in all measured automatic metrics and produces fewer terminology errors.
Hans Christian Andersen is one of the most famous Danish writers whose fairy tales have been read by almost every child. The first book of Andersen’s fairy tales appeared in Lithuanian in 1895 and their popularity has not faded ever since. However, most Lithuanian translations consist of fairy tales translated not from Danish, but from intermediate languages (English, Spanish, Italian, French). This article analyses abridged, retold versions and translations from the intermediate languages of Andersen’s tale “Thumbelina”. It is concluded that, due to commercialization and the desire to publish the tales more cheaply, they are distorted and simplified. As a result, the storyline changes and the artistically motivated, integral structural elements of content and form are violated.
The article presents the life and work of Bolesława Kopelówna, a Polish literary translator who was especially active (and widely criticised) in the interwar years in Poland, and is now almost completely forgotten. The article attempts to answer the following questions: why was Kopelówna so intensely criticised? Why has she disappeared from the collective memory? Why was she so active in the field of translation? And, no less crucially, who was this enigmatic figure of Bolesława Kopelówna? Through an application of microhistorical tools to fragments of Kopelówna’s life and work, I will re/deconstruct her seemingly non-existing archive. Combining interdisciplinary tools from literary history, history and feminist studies, my aim is not only to bring back the voice of a silenced, overlooked, and underestimated translator, but also to encourage other researchers to attempt to fill blank spaces in translation history.
Interpreters facilitate multi-lingual meetings but the affordable set of languages is often smaller than what is needed. Automatic simultaneous speech translation can extend the set of provided languages. We investigate if such an automatic system should rather follow the original speaker, or an interpreter to achieve better translation quality at the cost of increased delay. To answer the question, we release Europarl Simultaneous Interpreting Corpus (ESIC), 10 hours of recordings and transcripts of European Parliament speeches in English, with simultaneous interpreting into Czech and German. We evaluate quality and latency of speaker-based and interpreter-based spoken translation systems from English to Czech. We study the differences in implicit simplification and summarization of the human interpreter compared to a machine translation system trained to shorten the output to some extent. Finally, we perform human evaluation to measure information loss of each of these approaches.
A presente tradução é parte uma primeira revisão do trabalho tradutório apresentado em dissertação de mestrado (SOUZA, 2016). A proposta foi traduzir os dísticos elegíacos de Ovídio em uma forma que ecoasse o ritmo original do poema latino. Para isso, partiu-se da proposta de Carlos Alberto Nunes, a substituição de longas em posição princeps por tônicas. Porém, nesta tradução, ao contrário da de Nunes, permitiu-se também o seguimento desta tônica por apenas uma átona, formando troqueu que pode ser executado em performance como espondeu. Dessa maneira, a possibilidade de variação no metro foi mantida, mas os dátilos foram mantidos fixos no quinto pé do hexâmetro e segundo hemistíquio do pentâmetro. A cesura obrigatória do pentâmetro datílico foi executada com a aproximação de duas tônicas e enfatizada com espaçamento obrigatório, que induz o leitor ao reinício do ritmo. Para manter o andamento do metro, foram utilizados recursos como deslocamento de tônica, elisão entre o fim do hexâmetro e início do pentâmetro, elisões em geral. A presente revisão apresenta alterações no hexâmetro, que procura executar as suas cesuras, porém com o resultado de várias cesuras femininas. Além disso, o tom do poema procurou maior leveza e fluidez, com a eliminação de hipérbatos radicais e o rearranjo de informações dentro do dístico, que é a unidade dentro do poema. O poema traduzido, oitavo do livro 1, apresenta a figura da alcoviteira, personagem comum na comédia nova latina, que aqui é a presentada como uma bruxa que influencia a amada a extorquir os seus amantes e a desprezar o eu-poético, que sendo poeta, não tem como dar presentes valiosos além dos próprios poemas.
Language and Literature, Translating and interpreting
En este articulo se abordan las consecuencias del uso de la traducción automática neuronal entrenada a partir de corpus obtenidos de traducciones de géneros textuales específicos producidos en contextos profesionales concretos. Fenómenos como el translationese y el post-editese vinculados al uso de la traducción automática marcan la necesidad de orientar la investigación hacia nuevas maneras de abordar la calidad en la traducción profesional. Desde una perspectiva de salvaguarda de las lenguas en tanto que patrimonio cultural inmaterial, se plantean diversas cuestiones relacionadas con el efecto que puede tener, en las cultures y las lengua de llegada, el uso de la traducción automática, en términos de percepción y de calidad, y cómo su uso puede llegar a alterar el estándar de la lengua. Se apunta la necesidad de definir umbrales de calidad de la traducción que vayan más allá de los aspectos relacionados con la detección de errores, que es el tipo de correcciones de mayor viabilidad económica.
In this paper, we present DuTongChuan, a novel context-aware translation model for simultaneous interpreting. This model allows to constantly read streaming text from the Automatic Speech Recognition (ASR) model and simultaneously determine the boundaries of Information Units (IUs) one after another. The detected IU is then translated into a fluent translation with two simple yet effective decoding strategies: partial decoding and context-aware decoding. In practice, by controlling the granularity of IUs and the size of the context, we can get a good trade-off between latency and translation quality easily. Elaborate evaluation from human translators reveals that our system achieves promising translation quality (85.71% for Chinese-English, and 86.36% for English-Chinese), specially in the sense of surprisingly good discourse coherence. According to an End-to-End (speech-to-speech simultaneous interpreting) evaluation, this model presents impressive performance in reducing latency (to less than 3 seconds at most times). Furthermore, we successfully deploy this model in a variety of Baidu's products which have hundreds of millions of users, and we release it as a service in our AI platform.
In an effort to interpret black-box models, researches for developing explanation methods have proceeded in recent years. Most studies have tried to identify input pixels that are crucial to the prediction of a classifier. While this approach is meaningful to analyse the characteristic of blackbox models, it is also important to investigate pixels that interfere with the prediction. To tackle this issue, in this paper, we propose an explanation method that visualizes undesirable regions to classify an image as a target class. To be specific, we divide the concept of undesirable regions into two terms: (1) factors for a target class, which hinder that black-box models identify intrinsic characteristics of a target class and (2) factors for non-target classes that are important regions for an image to be classified as other classes. We visualize such undesirable regions on heatmaps to qualitatively validate the proposed method. Furthermore, we present an evaluation metric to provide quantitative results on ImageNet.
A main issue preventing the use of Convolutional Neural Networks (CNN) in end user applications is the low level of transparency in the decision process. Previous work on CNN interpretability has mostly focused either on localizing the regions of the image that contribute to the result or on building an external model that generates plausible explanations. However, the former does not provide any semantic information and the latter does not guarantee the faithfulness of the explanation. We propose an intermediate representation composed of multiple Semantically Interpretable Activation Maps (SIAM) indicating the presence of predefined attributes at different locations of the image. These attribute maps are then linearly combined to produce the final output. This gives the user insight into what the model has seen, where, and a final output directly linked to this information in a comprehensive and interpretable way. We test the method on the task of landscape scenicness (aesthetic value) estimation, using an intermediate representation of 33 attributes from the SUN Attributes database. The results confirm that SIAM makes it possible to understand what attributes in the image are contributing to the final score and where they are located. Since it is based on learning from multiple tasks and datasets, SIAM improve the explanability of the prediction without additional annotation efforts or computational overhead at inference time, while keeping good performances on both the final and intermediate tasks.