Cross-lingual Sentence Embedding using Multi-Task Learning

Sentence embeddings represent the meaning of a given sentence using a fixed dimensional vector. Different approaches have been proposed in the Natural Language Processing (NLP) community for learning encoders that can produce accurate sentence embeddings that perform well for diverse downstream tasks requiring sentence representations. Despite prior work focusing mainly on creating accurate sentence embeddings, how to keep private the sensitive information contained in the sentences remains an unexplored research problem. In this article, we propose Covering Metric Analytic Gaussian (CMAG), a covering metric Differential Privacy (DP) mechanism for sentence embeddings such that minimal random noise is added to a set of sentence embeddings produced by an encoder to protect the private information expressed in those sentences. Given a sentence embedding s , CMAG considers the Mahalanobis distance between s and the other sentence embeddings s ’ in the local neighbourhood of s to determine the minimal amount of random noise that must be added to s to obtain provable metric DP guarantees. Experimental results show that the proposed DP mechanism protects private information better than previously proposed DP mechanisms while reporting good performance in a broad range of downstream NLP tasks.

Background . Transformer-based models have become central to natural language processing, demonstrating state-of-the-art performance in semantic similarity assessment, a task critical for various applications. These models capture detailed relationships between text, advancing the ability to gauge semantic relatedness. Materials and Methods. The performance of sentence embedding models, including all-mpnet-base-v2 , all-MiniLM-L6-v2 , paraphrase-multilingual-mpnet-base-v2 , bge-base-en-v1.5 , all-roberta-large-v1 , a ll-distilroberta-v1 , LaBSE , paraphrase-MiniLM-L3-v2 , bge-large-en-v1.5 , was assessed across different dataset sizes with two datasets. The following preprocessing steps were applied to the datasets: lowercasing, removing stop words, cleaning from special symbols and numbers, and lemmatization. Cosine similarity scores with negative values, indicating semantic dissimilarity, were treated as equivalent to a human-annotated similarity score of 0, and non-negative cosine similarity values were scaled to the 0-5 range. Metrics such as R 2 , MSE, RMSE, MAE, Spearman’s Correlation Coefficient, and Kendall's Tau were used for evaluation. Results and Discussion. Models’ performance generally improves with increased data. Evaluation of sentence embedding models revealed performance variations. all-roberta-large-v1 showed strong accuracy with high R 2 values and low errors. BAAI/bge-large-en-v1.5 excelled in capturing semantic relationships, demonstrating high Spearman’s and Kendall's Tau coefficients. all-MiniLM-L6-v2 demonstrated the fastest embedding generation. BAAI/bge-base-en-v1.5 presented the lowest accuracy. Processing times generally increase with data size. Conclusion. This study highlights a trade-off between accuracy and efficiency in sentence embedding. Model selection depends on balancing these factors to align with specific application needs. In cases when requiring high accuracy should favor all-roberta-large-v1 , while those prioritizing speed would benefit from all-MiniLM-L6-v2 . BAAI/bge-large-en-v1.5 is most suitable for tasks demanding semantic understanding of text details. Keywords : semantic similarity, sentence embeddings, transformers.

Developing the utilized intelligent systems is increasingly important to learn effective text representations, especially extract the sentence features. Numerous previous studies have been concentrated on the task of sentence representation learning based on deep learning approaches. However, the present approaches are mostly proposed with the single task or replied on the labeled corpus when learning the embedding of the sentences. In this paper, we assess the factors in learning sentence representation and propose an efficient unsupervised learning framework with multi-task learning (USR-MTL), in which various text learning tasks are merged into the unitized framework. With the syntactic and semantic features of sentences, three different factors to some extent are reflected in the task of the sentence representation learning that is the wording, or the ordering of the neighbored sentences of a target sentence in other words. Hence, we integrate the word-order learning task, word prediction task, and the sentence-order learning task into the proposed framework to attain meaningful sentence embeddings. Here, the process of sentence embedding learning is reformulated as a multi-task learning framework of the sentence-level task and the two word-level tasks. Moreover, the proposed framework is motivated by an unsupervised learning algorithm utilizing the unlabeled corpus. Based on the experimental results, our approach achieves the state-of-the-art performances on the downstream natural language processing tasks compared to the popular unsupervised representation learning techniques. The experiments on representation visualization and task analysis demonstrate the effectiveness of the tasks in the proposed framework in creating reasonable sentence representations proving the capacity of the proposed unsupervised multi-task framework for the sentence representation learning.

This paper studies the relative importance of attention heads in Transformer-based models to aid their interpretability in cross-lingual and multi-lingual tasks. Prior research has found that only a few attention heads are important in each mono-lingual Natural Language Processing (NLP) task and pruning the remaining heads leads to comparable or improved performance of the model. However, the impact of pruning attention heads is not yet clear in cross-lingual and multi-lingual tasks. Through extensive experiments, we show that (1) pruning a number of attention heads in a multi-lingual Transformer-based model has, in general, positive effects on its performance in cross-lingual and multi-lingual tasks and (2) the attention heads to be pruned can be ranked using gradients and identified with a few trial experiments. Our experiments focus on sequence labeling tasks, with potential applicability on other cross-lingual and multi-lingual tasks. For comprehensiveness, we examine two pre-trained multi-lingual models, namely multi-lingual BERT (mBERT) and XLM-R, on three tasks across 9 languages each. We also discuss the validity of our findings and their extensibility to truly resource-scarce languages and other task settings.

The pursuit of universal, symmetric semantic representations within large language models (LLMs) faces a fundamental challenge: the inherent asymmetry of natural languages. Different languages exhibit vast disparities in syntactic structures, lexical choices, and cultural nuances, making the creation of a truly shared, symmetric embedding space a non-trivial task. This paper aims to address this critical problem by introducing a novel framework to forge robust and symmetric multilingual sentence embeddings. Our approach, named DACL (Dynamic Asymmetric Contrastive Learning), is anchored in two powerful asymmetric learning paradigms: Contrastive Learning and Dynamic Curriculum Learning (DCL). We extend Contrastive Learning to the multilingual context, where it asymmetrically treats semantically equivalent sentences from different languages (positive pairs) and sentences with distinct meanings (negative pairs) to enforce semantic symmetry in the target embedding space. To further refine this process, we incorporate Dynamic Curriculum Learning, which introduces a second layer of asymmetry by dynamically scheduling training instances from easy to hard. This dual-asymmetric strategy enables the model to progressively master complex cross-lingual relationships, starting with more obvious semantic equivalences and advancing to subtler ones. Our comprehensive experiments on benchmark cross-lingual tasks, including sentence retrieval and cross-lingual classification (XNLI, PAWS-X, MLDoc, MARC), demonstrate that DACL significantly outperforms a wide range of established baselines. The results validate our dual-asymmetric framework as a highly effective approach for forging robust multilingual embeddings, particularly excelling in tasks involving complex linguistic asymmetries. Ultimately, this work contributes a novel dual-asymmetric learning framework that effectively leverages linguistic asymmetry to achieve robust semantic symmetry across languages. It offers valuable insights for developing more capable, fair, and interpretable multilingual LLMs, emphasizing that deliberately leveraging asymmetry in the learning process is a highly effective strategy.

Multilingual knowledge graph (KG) embeddings provide latent semantic representations of entities and structured knowledge with cross-lingual inferences, which benefit various knowledge-driven cross-lingual NLP tasks. However, precisely learning such cross-lingual inferences is usually hindered by the low coverage of entity alignment in many KGs. Since many multilingual KGs also provide literal descriptions of entities, in this paper, we introduce an embedding-based approach which leverages a weakly aligned multilingual KG for semi-supervised cross-lingual learning using entity descriptions. Our approach performs co-training of two embedding models, i.e. a multilingual KG embedding model and a multilingual literal description embedding model. The models are trained on a large Wikipedia-based trilingual dataset where most entity alignment is unknown to training. Experimental results show that the performance of the proposed approach on the entity alignment task improves at each iteration of co-training, and eventually reaches a stage at which it significantly surpasses previous approaches. We also show that our approach has promising abilities for zero-shot entity alignment, and cross-lingual KG completion.

Word and sentence embeddings are useful feature representations in natural language processing. However, intrinsic evaluation for embeddings lags far behind, and there has been no significant update since the past decade. Word and sentence similarity tasks have become the de facto evaluation method. It leads models to overfit to such evaluations, negatively impacting embedding models' development. This paper first points out the problems using semantic similarity as the gold standard for word and sentence embedding evaluations. Further, we propose a new intrinsic evaluation method called EvalRank, which shows a much stronger correlation with downstream tasks. Extensive experiments are conducted based on 60+ models and popular datasets to certify our judgments. Finally, the practical evaluation toolkit is released for future benchmarking purposes.

Analysis of word embedding properties to inform their use in downstream NLP tasks has largely been studied by assessing nearest neighbors. However, geometric properties of the continuous feature space contribute directly to the use of embedding features in downstream models, and are largely unexplored. We consider four properties of word embedding geometry, namely: position relative to the origin, distribution of features in the vector space, global pairwise distances, and local pairwise distances. We define a sequence of transformations to generate new embeddings that expose subsets of these properties to downstream models and evaluate change in task performance to understand the contribution of each property to NLP models. We transform publicly available pretrained embeddings from three popular toolkits (word2vec, GloVe, and FastText) and evaluate on a variety of intrinsic tasks, which model linguistic information in the vector space, and extrinsic tasks, which use vectors as input to machine learning models. We find that intrinsic evaluations are highly sensitive to absolute position, while extrinsic tasks rely primarily on local similarity. Our findings suggest that future embedding models and post-processing techniques should focus primarily on similarity to nearby points in vector space.

In natural language processing (NLP) tasks, computing semantic textual similarity (STS) is crucial for capturing nuanced semantic differences in text. Traditional word vector methods, such as Word2Vec and GloVe, as well as deep learning models like BERT, face limitations in handling context dependency and polysemy and present challenges in computational resources and real-time processing. To address these issues, this paper introduces two novel methods. First, a sentence embedding generation method based on Kullback–Leibler Divergence (KLD) optimization is proposed, which enhances semantic differentiation between sentence vectors, thereby improving the accuracy of textual similarity computation. Second, this study proposes a framework incorporating RoBERTa knowledge distillation, which integrates the deep semantic insights of the RoBERTa model with prior methodologies to enhance sentence embeddings while preserving computational efficiency. Additionally, the study extends its contributions to sentiment analysis tasks by leveraging the enhanced embeddings for classification. The sentiment analysis experiments, conducted using a Stochastic Gradient Descent (SGD) classifier on the ACL IMDB dataset, demonstrate the effectiveness of the proposed methods, achieving high precision, recall, and F1 score metrics. To further augment model accuracy and efficacy, a feature selection approach is introduced, specifically through the Dynamic Principal Component Selection (DPCS) algorithm. The DPCS method autonomously identifies and prioritizes critical features, thus enriching the expressive capacity of sentence vectors and significantly advancing the accuracy of similarity computations. Experimental results demonstrate that our method outperforms existing methods in semantic similarity computation on the SemEval-2016 dataset. When evaluated using cosine similarity of average vectors, our model achieved a Pearson correlation coefficient (τ) of 0.470, a Spearman correlation coefficient (ρ) of 0.481, and a mean absolute error (MAE) of 2.100. Compared to traditional methods such as Word2Vec, GloVe, and FastText, our method significantly enhances similarity computation accuracy. Using TF-IDF-weighted cosine similarity evaluation, our model achieved a τ of 0.528, ρ of 0.518, and an MAE of 1.343. Additionally, in the cosine similarity assessment leveraging the Dynamic Principal Component Smoothing (DPCS) algorithm, our model achieved a τ of 0.530, ρ of 0.518, and an MAE of 1.320, further demonstrating the method’s effectiveness and precision in handling semantic similarity. These results indicate that our proposed method has high relevance and low error in semantic textual similarity tasks, thereby better capturing subtle semantic differences between texts.

In this study, we proposed a novel scale abbreviation method based on sentence embeddings and compared it to two established automatic scale abbreviation techniques. Scale abbreviation methods typically rely on administering the full scale to a large representative sample, which is often impractical in certain settings. Our approach leverages the semantic similarity among the items to select abbreviated versions of scales without requiring response data, offering a practical alternative for scale development. We found that the sentence embedding method performs comparably to the data-driven scale abbreviation approaches in terms of model fit, measurement accuracy, and ability estimates. In addition, our results reveal a moderate negative correlation between item discrimination parameters and semantic similarity indices, suggesting that semantically unique items may result in a higher discrimination power. This supports the notion that semantic features can be predictive of psychometric properties. However, this relationship was not observed for reverse-scored items, which may require further investigation. Overall, our findings suggest that the sentence embedding approach offers a promising solution for scale abbreviation, particularly in situations where large sample sizes are unavailable, and may eventually serve as an alternative to traditional data-driven methods.

In this paper, we present an approach to learn multilingual sentence embeddings using a bi-directional dual-encoder with additive margin softmax. The embeddings are able to achieve state-of-the-art results on the United Nations (UN) parallel corpus retrieval task. In all the languages tested, the system achieves P@1 of 86% or higher. We use pairs retrieved by our approach to train NMT models that achieve similar performance to models trained on gold pairs. We explore simple document-level embeddings constructed by averaging our sentence embeddings. On the UN document-level retrieval task, document embeddings achieve around 97% on P@1 for all experimented language pairs. Lastly, we evaluate the proposed model on the BUCC mining task. The learned embeddings with raw cosine similarity scores achieve competitive results compared to current state-of-the-art models, and with a second-stage scorer we achieve a new state-of-the-art level on this task.

BlendCSE: Blend contrastive learnings for sentence embeddings with rich semantics and transferability

The purpose of this study is to describe and evaluate a multilingual automated essay scoring (AES) system for grading essays in three languages. Two different sentence embedding models were evaluated within the AES system, multilingual BERT (mBERT) and language‐agnostic BERT sentence embedding (LaBSE). German, Italian, and Czech essays were holistically scored using the Common European Framework of Reference of Languages. The AES system with mBERT produced results that were consistent with human raters overall across all three language groups. The system also produced accurate predictions for some but not all of the score levels within each language. The AES system with LaBSE produced results that were even more consistent with the human raters overall across all three language groups compared to mBERT. In addition, the system produced accurate predictions for the majority of the score levels within each language. The performance differences between mBERT and LaBSE can be explained by considering how each language embedding model is implemented. Implications of this study for educational testing are also discussed.

Recent progress in task-oriented neural dialogue systems is largely focused on a handful of languages, as annotation of training data is tedious and expensive. Machine translation has been used to make systems multilingual, but this can introduce a pipeline of errors. Another promising solution is using cross-lingual transfer learning through pretrained multilingual models. Existing methods train multilingual models with additional code-mixed task data or refine the cross-lingual representations through parallel ontologies. In this work, we enhance the transfer learning process by intermediate fine-tuning of pretrained multilingual models, where the multilingual models are fine-tuned with different but related data and/or tasks. Specifically, we use parallel and conversational movie subtitles datasets to design cross-lingual intermediate tasks suitable for downstream dialogue tasks. We use only 200K lines of parallel data for intermediate fine-tuning which is already available for 1782 language pairs. We test our approach on the cross-lingual dialogue state tracking task for the parallel MultiWoZ (English -> Chinese, Chinese -> English) and Multilingual WoZ (English -> German, English -> Italian) datasets. We achieve impressive improvements (> 20% on joint goal accuracy) on the parallel MultiWoZ dataset and the Multilingual WoZ dataset over the vanilla baseline with only 10% of the target language task data and zero-shot setup respectively.

With growing applications of Machine Learning in daily lives Natural Language Processing (NLP) has emerged as a heavily researched area. Finding its applications in tasks ranging from simple Q/A chatbots to Fully fledged conversational AI, NLP models are vital. Word and Sentence embedding are one of the most common starting points of any NLP task. A word embedding represents a given word in a predefined vector-space while maintaining vector relations with similar or dis-similar entities. As such different pretrained embedding such as Word2Vec, GloVe, fasttext have been developed. These embedding generated on millions of words are however very large in terms of size. Having embedding with floating point precision also makes the downstream evaluation slow. In this paper we present a novel method to convert continuous embedding to its binary representation, thus reducing the overall size of the embedding while keeping the semantic and relational knowledge intact. This will facilitate an option of porting such big embedding onto devices where space is limited. We also present different approaches suitable for different downstream tasks based on the requirement of contextual and semantic information. Experiments have shown comparable result in downstream tasks with 7 to 15 times reduction in file size and about 5 % change in evaluation parameters.