Constantin Orǎsan

Leonardo Zilio, Hadeel Saadany, Prashant Sharma et al.

The detection and extraction of abbreviations from unstructured texts can help to improve the performance of Natural Language Processing tasks, such as machine translation and information retrieval. However, in terms of publicly available datasets, there is not enough data for training deep-neural-networks-based models to the point of generalising well over data. This paper presents PLOD, a large-scale dataset for abbreviation detection and extraction that contains 160k+ segments automatically annotated with abbreviations and their long forms. We performed manual validation over a set of instances and a complete automatic validation for this dataset. We then used it to generate several baseline models for detecting abbreviations and long forms. The best models achieved an F1-score of 0.92 for abbreviations and 0.89 for detecting their corresponding long forms. We release this dataset along with our code and all the models publicly in https://github.com/surrey-nlp/PLOD-AbbreviationDetection

Shenbin Qian, Constantin Orasan, Felix do Carmo et al.

In this paper, we focus on how current Machine Translation (MT) tools perform on the translation of emotion-loaded texts by evaluating outputs from Google Translate according to a framework proposed in this paper. We propose this evaluation framework based on the Multidimensional Quality Metrics (MQM) and perform a detailed error analysis of the MT outputs. From our analysis, we observe that about 50% of the MT outputs fail to preserve the original emotion. After further analysis of the errors, we find that emotion carrying words and linguistic phenomena such as polysemous words, negation, abbreviation etc., are common causes for these translation errors.

Hadeel Saadany, Catherine Breslin, Constantin Orăsan et al.

Transcription of legal proceedings is very important to enable access to justice. However, speech transcription is an expensive and slow process. In this paper we describe part of a combined research and industrial project for building an automated transcription tool designed specifically for the Justice sector in the UK. We explain the challenges involved in transcribing court room hearings and the Natural Language Processing (NLP) techniques we employ to tackle these challenges. We will show that fine-tuning a generic off-the-shelf pre-trained Automatic Speech Recognition (ASR) system with an in-domain language model as well as infusing common phrases extracted with a collocation detection model can improve not only the Word Error Rate (WER) of the transcribed hearings but avoid critical errors that are specific of the legal jargon and terminology commonly used in British courts.

Hadeel Saadany, Constantin Orasan, Emad Mohamed et al.

In the online world, Machine Translation (MT) systems are extensively used to translate User-Generated Text (UGT) such as reviews, tweets, and social media posts, where the main message is often the author's positive or negative attitude towards the topic of the text. However, MT systems still lack accuracy in some low-resource languages and sometimes make critical translation errors that completely flip the sentiment polarity of the target word or phrase and hence delivers a wrong affect message. This is particularly noticeable in texts that do not follow common lexico-grammatical standards such as the dialectical Arabic (DA) used on online platforms. In this research, we aim to improve the translation of sentiment in UGT written in the dialectical versions of the Arabic language to English. Given the scarcity of gold-standard parallel data for DA-EN in the UGT domain, we introduce a semi-supervised approach that exploits both monolingual and parallel data for training an NMT system initialised by a cross-lingual language model trained with supervised and unsupervised modeling objectives. We assess the accuracy of sentiment translation by our proposed system through a numerical 'sentiment-closeness' measure as well as human evaluation. We will show that our semi-supervised MT system can significantly help with correcting sentiment errors detected in the online translation of dialectical Arabic UGT.

Archchana Sindhujan, Girish A. Koushik, Shenbin Qian et al.

Quality Estimation (QE) aims to assess the quality of machine translation (MT) outputs without relying on reference translations, making it essential for real-world, large-scale MT evaluation. Large Language Models (LLMs) have shown significant promise in advancing the field of quality estimation of machine translation. However, most of the QE approaches solely rely on scalar quality scores, offering no explicit information about the translation errors that should drive these judgments. Moreover, for low-resource languages where annotated QE data is limited, existing approaches struggle to achieve reliable performance. To address these challenges, we introduce the first segment-level QE dataset for English to Malayalam, a severely resource-scarce language pair in the QE domain, comprising human-annotated Direct Assessment (DA) scores and Translation Quality Remarks (TQR), which are short, contextual, free-form annotator comments that describe translation errors. We further introduce ALOPE-RL, a policy-based reinforcement learning framework that trains efficient adapters based on policy rewards derived from DA score and TQR. Integrating error-aware rewards with ALOPE-RL, enables LLMs to reason about translation quality beyond numeric scores. Despite being trained on a small-scale QE dataset, ALOPE-RL achieves state-of-the-art performance on English to Malayalam QE using compact LLMs (<=4B parameters}) fine-tuned with LoRA and 4-bit quantization, outperforming both larger LLM-based baselines and leading encoder-based QE models. Our results demonstrate that error-aware, policy-based learning can deliver strong QE performance under limited data and compute budgets. We release our dataset, code, and trained models to support future research.

Tharindu Ranasinghe, Constantin Orasan, Ruslan Mitkov

Recent years have seen big advances in the field of sentence-level quality estimation (QE), largely as a result of using neural-based architectures. However, the majority of these methods work only on the language pair they are trained on and need retraining for new language pairs. This process can prove difficult from a technical point of view and is usually computationally expensive. In this paper we propose a simple QE framework based on cross-lingual transformers, and we use it to implement and evaluate two different neural architectures. Our evaluation shows that the proposed methods achieve state-of-the-art results outperforming current open-source quality estimation frameworks when trained on datasets from WMT. In addition, the framework proves very useful in transfer learning settings, especially when dealing with low-resourced languages, allowing us to obtain very competitive results.

Archchana Sindhujan, Diptesh Kanojia, Constantin Orasan et al.

This paper investigates the reference-less evaluation of machine translation for low-resource language pairs, known as quality estimation (QE). Segment-level QE is a challenging cross-lingual language understanding task that provides a quality score (0-100) to the translated output. We comprehensively evaluate large language models (LLMs) in zero/few-shot scenarios and perform instruction fine-tuning using a novel prompt based on annotation guidelines. Our results indicate that prompt-based approaches are outperformed by the encoder-based fine-tuned QE models. Our error analysis reveals tokenization issues, along with errors due to transliteration and named entities, and argues for refinement in LLM pre-training for cross-lingual tasks. We release the data, and models trained publicly for further research.

Jaleh Delfani, Constantin Orasan, Hadeel Saadany et al.

This study explores the use of Google Translate (GT) for translating mental healthcare (MHealth) information and evaluates its accuracy, comprehensibility, and implications for multilingual healthcare communication through analysing GT output in the MHealth domain from English to Persian, Arabic, Turkish, Romanian, and Spanish. Two datasets comprising MHealth information from the UK National Health Service website and information leaflets from The Royal College of Psychiatrists were used. Native speakers of the target languages manually assessed the GT translations, focusing on medical terminology accuracy, comprehensibility, and critical syntactic/semantic errors. GT output analysis revealed challenges in accurately translating medical terminology, particularly in Arabic, Romanian, and Persian. Fluency issues were prevalent across various languages, affecting comprehension, mainly in Arabic and Spanish. Critical errors arose in specific contexts, such as bullet-point formatting, specifically in Persian, Turkish, and Romanian. Although improvements are seen in longer-text translations, there remains a need to enhance accuracy in medical and mental health terminology and fluency, whilst also addressing formatting issues for a more seamless user experience. The findings highlight the need to use customised translation engines for Mhealth translation and the challenges when relying solely on machine-translated medical content, emphasising the crucial role of human reviewers in multilingual healthcare communication.

Hadeel Saadany, Ashraf Tantawy, Constantin Orasan

With the advent of Neural Machine Translation (NMT) systems, the MT output has reached unprecedented accuracy levels which resulted in the ubiquity of MT tools on almost all online platforms with multilingual content. However, NMT systems, like other state-of-the-art AI generative systems, are prone to errors that are deemed machine hallucinations. The problem with NMT hallucinations is that they are remarkably \textit{fluent} hallucinations. Since they are trained to produce grammatically correct utterances, NMT systems are capable of producing mistranslations that are too fluent to be recognised by both users of the MT tool, as well as by automatic quality metrics that are used to gauge their performance. In this paper, we introduce an authentic dataset of machine translation critical errors to point to the ethical and safety issues involved in the common use of MT. The dataset comprises mistranslations of Arabic mental health postings manually annotated with critical error types. We also show how the commonly used quality metrics do not penalise critical errors and highlight this as a critical issue that merits further attention from researchers.

Anastasiia Bezobrazova, Miriam Seghiri, Constantin Orasan

This paper compares the accuracy of the terms extracted using SketchEngine, TBXTools and ChatGPT. In addition, it evaluates the quality of the definitions produced by ChatGPT for these terms. The research is carried out on a comparable corpus of fashion magazines written in English and Russian collected from the web. A gold standard for the fashion terminology was also developed by identifying web pages that can be harvested automatically and contain definitions of terms from the fashion domain in English and Russian. This gold standard was used to evaluate the quality of the extracted terms and of the definitions produced. Our evaluation shows that TBXTools and SketchEngine, while capable of high recall, suffer from reduced precision as the number of terms increases, which affects their overall performance. Conversely, ChatGPT demonstrates superior performance, maintaining or improving precision as more terms are considered. Analysis of the definitions produced by ChatGPT for 60 commonly used terms in English and Russian shows that ChatGPT maintains a reasonable level of accuracy and fidelity across languages, but sometimes the definitions in both languages miss crucial specifics and include unnecessary deviations. Our research reveals that no single tool excels universally; each has strengths suited to particular aspects of terminology extraction and application.

Hadeel Saadany, Swapnil Bhosale, Samarth Agrawal et al.

This paper addresses the challenge of improving user experience on e-commerce platforms by enhancing product ranking relevant to users' search queries. Ambiguity and complexity of user queries often lead to a mismatch between the user's intent and retrieved product titles or documents. Recent approaches have proposed the use of Transformer-based models, which need millions of annotated query-title pairs during the pre-training stage, and this data often does not take user intent into account. To tackle this, we curate samples from existing datasets at eBay, manually annotated with buyer-centric relevance scores and centrality scores, which reflect how well the product title matches the users' intent. We introduce a User-intent Centrality Optimization (UCO) approach for existing models, which optimises for the user intent in semantic product search. To that end, we propose a dual-loss based optimisation to handle hard negatives, i.e., product titles that are semantically relevant but do not reflect the user's intent. Our contributions include curating challenging evaluation sets and implementing UCO, resulting in significant product ranking efficiency improvements observed for different evaluation metrics. Our work aims to ensure that the most buyer-centric titles for a query are ranked higher, thereby, enhancing the user experience on e-commerce platforms.

Archchana Sindhujan, Shenbin Qian, Chan Chi Chun Matthew et al.

Large Language Models (LLMs) have shown remarkable performance across a wide range of natural language processing tasks. Quality Estimation (QE) for Machine Translation (MT), which assesses the quality of a source-target pair without relying on reference translations, remains a challenging cross-lingual task for LLMs. The challenges stem from the inherent limitations of existing LLM-based QE systems, which are pre-trained for causal language modelling rather than regression-specific tasks, further elevated by the presence of low-resource languages given pre-training data distribution. This paper introduces ALOPE, an adaptive layer-optimization framework designed to enhance LLM-based QE by restructuring Transformer representations through layer-wise adaptation for improved regression-based prediction. Our framework integrates low-rank adapters (LoRA) with regression task heads, leveraging selected pre-trained Transformer layers for improved cross-lingual alignment. In addition to the layer-specific adaptation, ALOPE introduces two strategies-dynamic weighting, which adaptively combines representations from multiple layers, and multi-head regression, which aggregates regression losses from multiple heads for QE. Our framework shows improvements over various existing LLM-based QE approaches. Empirical evidence suggests that intermediate Transformer layers in LLMs provide contextual representations that are more aligned with the cross-lingual nature of the QE task. We make resultant models and framework code publicly available for further research, also allowing existing LLM-based MT frameworks to be scaled with QE capabilities.

Shenbin Qian, Diptesh Kanojia, Samarth Agrawal et al.

E-commerce information retrieval (IR) systems struggle to simultaneously achieve high accuracy in interpreting complex user queries and maintain efficient processing of vast product catalogs. The dual challenge lies in precisely matching user intent with relevant products while managing the computational demands of real-time search across massive inventories. In this paper, we propose a Nested Embedding Approach to product Retrieval and Ranking, called NEAR$^2$, which can achieve up to $12$ times efficiency in embedding size at inference time while introducing no extra cost in training and improving performance in accuracy for various encoder-based Transformer models. We validate our approach using different loss functions for the retrieval and ranking task, including multiple negative ranking loss and online contrastive loss, on four different test sets with various IR challenges such as short and implicit queries. Our approach achieves an improved performance over a smaller embedding dimension, compared to any existing models.

Shenbin Qian, Constantin Orăsan, Diptesh Kanojia et al.

Evaluating machine translation (MT) of user-generated content (UGC) involves unique challenges such as checking whether the nuance of emotions from the source are preserved in the target text. Recent studies have proposed emotion-related datasets, frameworks and models to automatically evaluate MT quality of Chinese UGC, without relying on reference translations. However, whether these models are robust to the challenge of preserving emotional nuances has been left largely unexplored. To address this gap, we introduce a novel method inspired by information theory which generates challenging Chinese homophone words related to emotions, by leveraging the concept of self-information. Our approach generates homophones that were observed to cause translation errors in emotion preservation, and exposes vulnerabilities in MT systems and their evaluation methods when tackling emotional UGC. We evaluate the efficacy of our method using human evaluation for the quality of these generated homophones, and compare it with an existing one, showing that our method achieves higher correlation with human judgments. The generated Chinese homophones, along with their manual translations, are utilized to generate perturbations and to probe the robustness of existing quality evaluation models, including models trained using multi-task learning, fine-tuned variants of multilingual language models, as well as large language models (LLMs). Our results indicate that LLMs with larger size exhibit higher stability and robustness to such perturbations. We release our data and code for reproducibility and further research.

Shiyi Tan, Constantin Orăsan, Sabine Braun

This paper reports on the results from a pilot study investigating the impact of automatic speech recognition (ASR) technology on interpreting quality in remote healthcare interpreting settings. Employing a within-subjects experiment design with four randomised conditions, this study utilises scripted medical consultations to simulate dialogue interpreting tasks. It involves four trainee interpreters with a language combination of Chinese and English. It also gathers participants' experience and perceptions of ASR support through cued retrospective reports and semi-structured interviews. Preliminary data suggest that the availability of ASR, specifically the access to full ASR transcripts and to ChatGPT-generated summaries based on ASR, effectively improved interpreting quality. Varying types of ASR output had different impacts on the distribution of interpreting error types. Participants reported similar interactive experiences with the technology, expressing their preference for full ASR transcripts. This pilot study shows encouraging results of applying ASR to dialogue-based healthcare interpreting and offers insights into the optimal ways to present ASR output to enhance interpreter experience and performance. However, it should be emphasised that the main purpose of this study was to validate the methodology and that further research with a larger sample size is necessary to confirm these findings.

Prashant Sharma, Hadeel Saadany, Leonardo Zilio et al.

Acronyms are abbreviated units of a phrase constructed by using initial components of the phrase in a text. Automatic extraction of acronyms from a text can help various Natural Language Processing tasks like machine translation, information retrieval, and text summarisation. This paper discusses an ensemble approach for the task of Acronym Extraction, which utilises two different methods to extract acronyms and their corresponding long forms. The first method utilises a multilingual contextual language model and fine-tunes the model to perform the task. The second method relies on a convolutional neural network architecture to extract acronyms and append them to the output of the previous method. We also augment the official training dataset with additional training samples extracted from several open-access journals to help improve the task performance. Our dataset analysis also highlights the noise within the current task dataset. Our approach achieves the following macro-F1 scores on test data released with the task: Danish (0.74), English-Legal (0.72), English-Scientific (0.73), French (0.63), Persian (0.57), Spanish (0.65), Vietnamese (0.65). We release our code and models publicly.

Hadeel Saadany, Constantin Orasan, Emad Mohamed et al.

In translating text where sentiment is the main message, human translators give particular attention to sentiment-carrying words. The reason is that an incorrect translation of such words would miss the fundamental aspect of the source text, i.e. the author's sentiment. In the online world, MT systems are extensively used to translate User-Generated Content (UGC) such as reviews, tweets, and social media posts, where the main message is often the author's positive or negative attitude towards the topic of the text. It is important in such scenarios to accurately measure how far an MT system can be a reliable real-life utility in transferring the correct affect message. This paper tackles an under-recognised problem in the field of machine translation evaluation which is judging to what extent automatic metrics concur with the gold standard of human evaluation for a correct translation of sentiment. We evaluate the efficacy of conventional quality metrics in spotting a mistranslation of sentiment, especially when it is the sole error in the MT output. We propose a numerical `sentiment-closeness' measure appropriate for assessing the accuracy of a translated affect message in UGC text by an MT system. We will show that incorporating this sentiment-aware measure can significantly enhance the correlation of some available quality metrics with the human judgement of an accurate translation of sentiment.

Hadeel Saadany, Constantin Orasan

Social media companies as well as authorities make extensive use of artificial intelligence (AI) tools to monitor postings of hate speech, celebrations of violence or profanity. Since AI software requires massive volumes of data to train computers, Machine Translation (MT) of the online content is commonly used to process posts written in several languages and hence augment the data needed for training. However, MT mistakes are a regular occurrence when translating sentiment-oriented user-generated content (UGC), especially when a low-resource language is involved. The adequacy of the whole process relies on the assumption that the evaluation metrics used give a reliable indication of the quality of the translation. In this paper, we assess the ability of automatic quality metrics to detect critical machine translation errors which can cause serious misunderstanding of the affect message. We compare the performance of three canonical metrics on meaningless translations where the semantic content is seriously impaired as compared to meaningful translations with a critical error which exclusively distorts the sentiment of the source text. We conclude that there is a need for fine-tuning of automatic metrics to make them more robust in detecting sentiment critical errors.

Diptesh Kanojia, Marina Fomicheva, Tharindu Ranasinghe et al.

Current Machine Translation (MT) systems achieve very good results on a growing variety of language pairs and datasets. However, they are known to produce fluent translation outputs that can contain important meaning errors, thus undermining their reliability in practice. Quality Estimation (QE) is the task of automatically assessing the performance of MT systems at test time. Thus, in order to be useful, QE systems should be able to detect such errors. However, this ability is yet to be tested in the current evaluation practices, where QE systems are assessed only in terms of their correlation with human judgements. In this work, we bridge this gap by proposing a general methodology for adversarial testing of QE for MT. First, we show that despite a high correlation with human judgements achieved by the recent SOTA, certain types of meaning errors are still problematic for QE to detect. Second, we show that on average, the ability of a given model to discriminate between meaning-preserving and meaning-altering perturbations is predictive of its overall performance, thus potentially allowing for comparing QE systems without relying on manual quality annotation.

Hadeel Saadany, Constantin Orasan, Rocio Caro Quintana et al.

Although emotions are universal concepts, transferring the different shades of emotion from one language to another may not always be straightforward for human translators, let alone for machine translation systems. Moreover, the cognitive states are established by verbal explanations of experience which is shaped by both the verbal and cultural contexts. There are a number of verbal contexts where expression of emotions constitutes the pivotal component of the message. This is particularly true for User-Generated Content (UGC) which can be in the form of a review of a product or a service, a tweet, or a social media post. Recently, it has become common practice for multilingual websites such as Twitter to provide an automatic translation of UGC to reach out to their linguistically diverse users. In such scenarios, the process of translating the user's emotion is entirely automatic with no human intervention, neither for post-editing nor for accuracy checking. In this research, we assess whether automatic translation tools can be a successful real-life utility in transferring emotion in user-generated multilingual data such as tweets. We show that there are linguistic phenomena specific of Twitter data that pose a challenge in translation of emotions in different languages. We summarise these challenges in a list of linguistic features and show how frequent these features are in different language pairs. We also assess the capacity of commonly used methods for evaluating the performance of an MT system with respect to the preservation of emotion in the source text.

Tharindu Ranasinghe, Constantin Orasan, Ruslan Mitkov

Most studies on word-level Quality Estimation (QE) of machine translation focus on language-specific models. The obvious disadvantages of these approaches are the need for labelled data for each language pair and the high cost required to maintain several language-specific models. To overcome these problems, we explore different approaches to multilingual, word-level QE. We show that these QE models perform on par with the current language-specific models. In the cases of zero-shot and few-shot QE, we demonstrate that it is possible to accurately predict word-level quality for any given new language pair from models trained on other language pairs. Our findings suggest that the word-level QE models based on powerful pre-trained transformers that we propose in this paper generalise well across languages, making them more useful in real-world scenarios.

Hadeel Saadany, Emad Mohamed, Constantin Orasan

One very common type of fake news is satire which comes in a form of a news website or an online platform that parodies reputable real news agencies to create a sarcastic version of reality. This type of fake news is often disseminated by individuals on their online platforms as it has a much stronger effect in delivering criticism than through a straightforward message. However, when the satirical text is disseminated via social media without mention of its source, it can be mistaken for real news. This study conducts several exploratory analyses to identify the linguistic properties of Arabic fake news with satirical content. We exploit these features to build a number of machine learning models capable of identifying satirical fake news with an accuracy of up to 98.6%.

Hadeel Saadany, Constantin Orasan

Since the advent of Neural Machine Translation (NMT) approaches there has been a tremendous improvement in the quality of automatic translation. However, NMT output still lacks accuracy in some low-resource languages and sometimes makes major errors that need extensive post-editing. This is particularly noticeable with texts that do not follow common lexico-grammatical standards, such as user generated content (UGC). In this paper we investigate the challenges involved in translating book reviews from Arabic into English, with particular focus on the errors that lead to incorrect translation of sentiment polarity. Our study points to the special characteristics of Arabic UGC, examines the sentiment transfer errors made by Google Translate of Arabic UGC to English, analyzes why the problem occurs, and proposes an error typology specific of the translation of Arabic UGC. Our analysis shows that the output of online translation tools of Arabic UGC can either fail to transfer the sentiment at all by producing a neutral target text, or completely flips the sentiment polarity of the target word or phrase and hence delivers a wrong affect message. We address this problem by fine-tuning an NMT model with respect to sentiment polarity showing that this approach can significantly help with correcting sentiment errors detected in the online translation of Arabic UGC.

Tharindu Ranasinghe, Alistair Plum, Constantin Orasan et al.

This paper presents the RGCL team submission to SemEval 2020 Task 6: DeftEval, subtasks 1 and 2. The system classifies definitions at the sentence and token levels. It utilises state-of-the-art neural network architectures, which have some task-specific adaptations, including an automatically extended training set. Overall, the approach achieves acceptable evaluation scores, while maintaining flexibility in architecture selection.

Tharindu Ranasinghe, Constantin Orasan, Ruslan Mitkov

This paper presents the team TransQuest's participation in Sentence-Level Direct Assessment shared task in WMT 2020. We introduce a simple QE framework based on cross-lingual transformers, and we use it to implement and evaluate two different neural architectures. The proposed methods achieve state-of-the-art results surpassing the results obtained by OpenKiwi, the baseline used in the shared task. We further fine tune the QE framework by performing ensemble and data augmentation. Our approach is the winning solution in all of the language pairs according to the WMT 2020 official results.

Tharindu Ranasinghe, Constantin Orasan, Ruslan Mitkov

Matching and retrieving previously translated segments from a Translation Memory is the key functionality in Translation Memories systems. However this matching and retrieving process is still limited to algorithms based on edit distance which we have identified as a major drawback in Translation Memories systems. In this paper we introduce sentence encoders to improve the matching and retrieving process in Translation Memories systems - an effective and efficient solution to replace edit distance based algorithms.