Textual Similarity Task Research Articles

Slovak morphological tokenizer using the Byte-Pair Encoding algorithm

This study introduces a new approach to text tokenization, SlovaK Morphological Tokenizer (SKMT), which integrates the morphology of the Slovak language into the training process using the Byte-Pair Encoding (BPE) algorithm. Unlike conventional tokenizers, SKMT focuses on preserving the integrity of word roots in individual tokens, crucial for maintaining lexical meaning. The methodology involves segmenting and extracting word roots from morphological dictionaries and databases, followed by corpus preprocessing and training SKMT alongside a traditional BPE tokenizer. Comparative evaluation against existing tokenizers demonstrates SKMT’s outstanding ability to maintain root integrity, achieving 99.7% root integrity compared to SlovakBERT (90.5%) and a pureBPE tokenizer (93.1%). Further validation involved fine-tuning models on a sentiment classification NLP task, where models trained with SKMT achieved an F1-score improvement of 3.5% over those trained with conventional BPE tokenization, followed by a focus on the Semantic Textual Similarity (STS) task. These findings suggest that training language models on the SKMT tokenizer significantly enhances model performance and quality.

Soft cosine and extended cosine adaptation for pre-trained language model semantic vector analysis

Semantic textual analysis is a natural language processing task that has enjoyed several research contributions towards solving diverse real-life problems. Vector comparison is a core subtask in semantic textual similarity analysis. A plethora of solutions including recent state-of-the-art transformer-based pre-trained language models for transfer learning have focused on using only cosine similarity for embedding evaluation in downstream tasks and ignored other vector comparison methods. To investigate the relative performance of some such ignored measures, this work proposes novel adaptations for soft cosine and extended cosine vector measures. We investigate their performance against the conventional cosine measure, distance-weighted cosine, vector similarity measure, negative Manhattan, and Euclidean distances on downstream semantic textual similarity tasks, under same conditions, for the first time in literature. Adopting transformer-based Universal sentence encoder, SBERT, SRoBERTa, SimCSE, and ST5 for text encoding; the performances of the adapted measures are evaluated on diverse real world datasets using Pearson, Spearman, accuracy and F1 evaluation metrics. Results obtained show that the adapted measures significantly surpass previously reported state-of-the-art cosine similarity-based correlations in several test cases considered.

Simple Data Transformations for Mitigating the Syntactic Similarity to Improve Sentence Embeddings at Supervised Contrastive Learning

Contrastive learning of sentence representations has achieved great improvements in several natural language processing tasks. However, the supervised contrastive learning model trained on the natural language inference (NLI) dataset is insufficient to elucidate the semantics of sentences since it is prone to make a prediction based on heuristics. Herein, by using the ParsEVAL and the word overlap metric, it is shown that sentence pairs in the NLI dataset have strong syntactic similarity and propose a framework to compensate for this problem in two aspects. 1) Apply simple syntactic transformations to the hypothesis and 2) expand the objective to SupCon Loss to leverage variants of sentences. The method is evaluated on semantic textual similarity (STS) tasks and transfer tasks. The proposed methods improve the performance of the BERT‐based baseline in STS Benchmark and SICK Relatedness by 1.48% and 2.2%. Furthermore, the model achieves 82.65% on the HANS benchmark dataset, to the best of our knowledge, which is a state‐of‐the‐art performance demonstrating that our approach is effective in grasping semantics without heuristics in the NLI dataset at supervised contrastive learning. The code is available at https://github.com/whnhch/Break-the-Similarity.

SEBGM: Sentence Embedding Based on Generation Model with multi-task learning

Sentence embedding, which aims to learn an effective representation of a sentence, is a significant part for downstream tasks. Recently, using contrastive learning and pre-trained model, most methods of sentence embedding achieve encouraging results. However, on the one hand, these methods utilize discrete data augmentation to obtain positive samples performing contrastive learning, which could distort the original semantic of sentences. On the other hand, most methods directly employ the contrastive frameworks of computer vision to perform contrastive learning, which could confine the contrastive training due to the discrete and sparse text data compared with image data. To solve the issues above, we design a novel contrastive framework based on generation model with multi-task learning by supervised contrastive training on the dataset of natural language inference (NLI) to obtain meaningful sentence embedding (SEBGM). SEBGM makes use of multi-task learning to enhance the usage of word-level and sentence-level semantic information of samples. In this way, the positive samples of SEBGM are from NLI rather than data augmentation. Extensive experiments show that our proposed SEBGM can advance the state-of-the-art sentence embedding on the semantic textual similarity (STS) tasks by utilizing multi-task learning.

DenoSent: A Denoising Objective for Self-Supervised Sentence Representation Learning

Contrastive-learning-based methods have dominated sentence representation learning. These methods regularize the representation space by pulling similar sentence representations closer and pushing away the dissimilar ones and have been proven effective in various NLP tasks, e.g., semantic textual similarity (STS) tasks. However, it is challenging for these methods to learn fine-grained semantics as they only learn from the inter-sentence perspective, i.e., their supervision signal comes from the relationship between data samples. In this work, we propose a novel denoising objective that inherits from another perspective, i.e., the intra-sentence perspective. By introducing both discrete and continuous noise, we generate noisy sentences and then train our model to restore them to their original form. Our empirical evaluations demonstrate that this approach delivers competitive results on both semantic textual similarity (STS) and a wide range of transfer tasks, standing up well in comparison to contrastive-learning-based methods. Notably, the proposed intra-sentence denoising objective complements existing inter-sentence contrastive methodologies and can be integrated with them to further enhance performance. Our code is available at https://github.com/xinghaow99/DenoSent.

Contrastive sentence representation learning with adaptive false negative cancellation

Contrastive sentence representation learning has made great progress thanks to a range of text augmentation strategies and hard negative sampling techniques. However, most studies directly employ in-batch samples as negative samples, ignoring the semantic relationship between negative samples and anchors, which may lead to negative sampling bias. To address this issue, we propose similarity and relative-similarity strategies for identifying potential false negatives. Moreover, we introduce adaptive false negative elimination and attraction methods to mitigate their adverse effects. Our proposed approaches can also be considered semi-supervised contrastive learning, as the identified false negatives can be viewed as either negative or positive samples for contrastive learning in adaptive false negative elimination and attraction methods. By fusing information from positive and negative pairs, contrastive learning learns rich and discriminative representations that capture the intrinsic characteristics of the sentence. Experimental results indicate that our proposed strategies and methods can bring further significant performance improvements. Specifically, the combination of similarity strategy and adaptive false negative elimination method achieves the best results, yielding an average performance gain of 2.1% compared to SimCSE in semantic textual similarity (STS) tasks. Furthermore, our approach is generalizable and can be applied to different text data augmentation strategies and certain existing contrastive sentence representation learning models. Our experimental code and data are publicly available at the link: https://github.com/Linda230/AFNC.

Grouped Contrastive Learning of Self-Supervised Sentence Representation

This paper proposes a method called Grouped Contrastive Learning of self-supervised Sentence Representation (GCLSR), which can learn an effective and meaningful representation of sentences. Previous works maximize the similarity between two vectors to be the objective of contrastive learning, suffering from the high-dimensionality of the vectors. In addition, most previous works have adopted discrete data augmentation to obtain positive samples and have directly employed a contrastive framework from computer vision to perform contrastive training, which could hamper contrastive training because text data are discrete and sparse compared with image data. To solve these issues, we design a novel framework of contrastive learning, i.e., GCLSR, which divides the high-dimensional feature vector into several groups and respectively computes the groups’ contrastive losses to make use of more local information, eventually obtaining a more fine-grained sentence representation. In addition, in GCLSR, we design a new self-attention mechanism and both a continuous and a partial-word vector augmentation (PWVA). For the discrete and sparse text data, the use of self-attention could help the model focus on the informative words by measuring the importance of every word in a sentence. By using the PWVA, GCLSR can obtain high-quality positive samples used for contrastive learning. Experimental results demonstrate that our proposed GCLSR achieves an encouraging result on the challenging datasets of the semantic textual similarity (STS) task and transfer task.

Refined SBERT: Representing sentence BERT in manifold space

Distributed sentence representations have shown great power in a wide range of natural language processing (NLP) tasks. Meanwhile, a contextualized sentence representation, called Sentence BERT (SBERT), achieves excellent performance in quite a few NLP tasks. However, Sentence BERT is typically learned in the Euclidean space, the geometric structure of sentence representations and their relations to the representations of sentence’s contexts has not been carefully studied yet. In this paper, we propose a new sentence representation method, named Refined SBERT, which utilizes manifold learning to refine sentence BERT by re-embedding sentence vectors from the original embedding space to a new refined semantic space. In order to map sentences to manifold space, we utilize neighborhood preserving embedding to construct the local manifold structure of the sentences. Our method can discover the local geometric structure and obtain a compact sentence BERT subspace, which can best detect the essential semantic structure. We conduct comprehensive experiments on various sentence embedding tasks including semantic textual similarity tasks, text classification and document clustering, and the experimental results show that the proposed model achieved promising results, comparing to several popular sentence representations.

Contrastive Learning Models for Sentence Representations

Sentence representation learning is a crucial task in natural language processing, as the quality of learned representations directly influences downstream tasks, such as sentence classification and sentiment analysis. Transformer-based pretrained language models such as bidirectional encoder representations from transformers (BERT) have been extensively applied to various natural language processing tasks, and have exhibited moderately good performance. However, the anisotropy of the learned embedding space prevents BERT sentence embeddings from achieving good results in the semantic textual similarity tasks. It has been shown that contrastive learning can alleviate the anisotropy problem and significantly improve sentence representation performance. Therefore, there has been a surge in the development of models that utilize contrastive learning to fine-tune BERT-like pretrained language models to learn sentence representations. But no systematic review of contrastive learning models for sentence representations has been conducted. To fill this gap, this article summarizes and categorizes the contrastive learning based sentence representation models, common evaluation tasks for assessing the quality of learned representations, and future research directions. Furthermore, we select several representative models for exhaustive experiments to illustrate the quantitative improvement of various strategies on sentence representations.

Parameter-efficient feature-based transfer for paraphrase identification

Abstract There are many types of approaches for Paraphrase Identification (PI), an NLP task of determining whether a sentence pair has equivalent semantics. Traditional approaches mainly consist of unsupervised learning and feature engineering, which are computationally inexpensive. However, their task performance is moderate nowadays. To seek a method that can preserve the low computational costs of traditional approaches but yield better task performance, we take an investigation into neural network-based transfer learning approaches. We discover that by improving the usage of parameters efficiently for feature-based transfer, our research goal can be accomplished. Regarding the improvement, we propose a pre-trained task-specific architecture. The fixed parameters of the pre-trained architecture can be shared by multiple classifiers with small additional parameters. As a result, the computational cost left involving parameter update is only generated from classifier-tuning: the features output from the architecture combined with lexical overlap features are fed into a single classifier for tuning. Furthermore, the pre-trained task-specific architecture can be applied to natural language inference and semantic textual similarity tasks as well. Such technical novelty leads to slight consumption of computational and memory resources for each task and is also conducive to power-efficient continual learning. The experimental results show that our proposed method is competitive with adapter-BERT (a parameter-efficient fine-tuning approach) over some tasks while consuming only 16% trainable parameters and saving 69-96% time for parameter update.

A Siamese Neural Network for Learning Semantically-Informed Sentence Embeddings

Semantic representation is a way of expressing the meaning of a text that can be processed by a machine to serve a particular natural language processing (NLP) task that usually requires meaning comprehension such as text summarisation, question answering or machine translation. In this paper, we present a semantic parsing model based on neural networks to obtain semantic representation of a given sentence. We utilise semantic representation of each sentence to generate semantically informed sentence embeddings for extrinsic evaluation of the proposed semantic parser, in particular for the semantic textual similarity task. Our neural parser utilises self-attention mechanism to learn semantic relations between words in a sentence to generate semantic representation of a sentence in UCCA (Universal Conceptual Cognitive Annotation) semantic annotation framework (Abend and Rappoport, 2013), which is a cross-linguistically applicable graph-based semantic representation. The UCCA representations are conveyed into a Siamese Neural Network built on top of two Recursive Neural Networks (Siamese-RvNN) to derive semantically informed sentence embeddings which are evaluated on semantic textual similarity task. We conduct both single-lingual and cross-lingual experiments with zero-shot and few-shot learning, which have shown superior performance even in low-resource scenario. The experimental results show that the proposed self-attentive neural parser outperforms the other parsers in the literature on English and German, and shows significant improvement in the cross-lingual setting for French which has comparatively low sources. Moreover, the results obtained from other downstream tasks such as sentiment analysis confirm that semantically informed sentence embeddings provide higher-quality embeddings compared to other pre-trained models such as SBERT (Reimers et al., 2019) or SimCSE (Gao et al., 2021), which do not utilise such structured information.

SGL: Symbolic Goal Learning in a Hybrid, Modular Framework for Human Instruction Following

This paper investigates human instruction following for robotic manipulation via a hybrid, modular system with symbolic and connectionist elements. Symbolic methods build modular systems with semantic parsing and task planning modules for producing sequences of actions from natural language requests. Modern connectionist methods employ deep neural networks that learn visual and linguistic features for mapping inputs to a sequence of low-level actions, in an end-to-end fashion. The hybrid, modular system blends these two approaches to create a modular framework: it formulates instruction following as symbolic goal learning via deep neural networks followed by task planning via symbolic planners. Connectionist and symbolic modules are bridged with Planning Domain Definition Language. The vision-and-language learning network predicts its goal representation, which is sent to a planner for producing a task-completing action sequence. For improving the flexibility of natural language, we further incorporate implicit human intents with explicit human instructions. To learn generic features for vision and language, we propose to separately pretrain vision and language encoders on scene graph parsing and semantic textual similarity tasks. Benchmarking evaluates the impacts of different components of, or options for, the vision-and-language learning model and shows the effectiveness of pretraining strategies. Manipulation experiments conducted in the simulator AI2THOR show the robustness of the framework to novel scenarios.

A novel locality-sensitive hashing relational graph matching network for semantic textual similarity measurement

Recent efforts adopt interaction-based models to construct the interaction of words between sentences, which aim to predict whether two sentences are semantically equivalent or not in semantic textual similarity (STS) task. However, these methods lack the global semantic awareness, which make it difficult to distinguish syntactic differences and also suffer from the inference time cost, primarily due to the calculation of the pair-interactions of words. A novel model called Locality-Sensitive Hashing Relational Graph Matching Network (LSHRGMN) is therefore proposed, which tackles these problems by syntactic dependency graph and locality-sensitive hashing (LSH). Specifically, syntactic dependency graph is aware of the global semantic information via rooting in each word to construct several trees and merging all the trees into one graph. LSH mechanism is introduced into pair-interactions of words for the inference efficiency problem. Extensive experiments are conducted on three real-world datasets, and the result shows that the proposed approach acquires higher accuracy and intriguing inference speed.

A Quantum Language-Inspired Tree Structural Text Representation for Semantic Analysis

Text representation is an important topic in the field of natural language processing, which can effectively transfer knowledge to downstream tasks. To extract effective semantic information from text with unsupervised methods, this paper proposes a quantum language-inspired tree structural text representation model to study the correlations between words with variable distance for semantic analysis. Combining the different semantic contributions of associated words in different syntax trees, a syntax tree-based attention mechanism is established to highlight the semantic contributions of non-adjacent associated words and weaken the semantic weight of adjacent non-associated words. Moreover, the tree-based attention mechanism includes not only the overall information of entangled words in the dictionary but also the local grammatical structure of word combinations in different sentences. Experimental results on semantic textual similarity tasks show that the proposed method obtains significant performances over the state-of-the-art sentence embeddings.

TA-SBERT: Token Attention Sentence-BERT for Improving Sentence Representation

A sentence embedding vector can be obtained by connecting a global average pooling (GAP) to a pre-trained language model. The problem of such a sentence embedding vector using a GAP is that it is generated with the same weight for all words appearing in the sentence. We propose a novel sentence embedding-method-based model Token Attention-SentenceBERT (TA-SBERT) to address this problem. The rationale of TA-SBERT is to enhance the performance of sentence embedding by introducing three strategies. First, we convert the base form while preprocessing the input sentence to reduce misunderstanding. Second, we propose a novel Token Attention (TA) technique that distinguishes important words to produce more informative sentence vectors. Third, we increase stability of fine-tuning to avoid catastrophic forgetting by adding a reconstruction loss to the word embedding vector. Extensive ablation studies demonstrate that our TA-SBERT outperforms the original SentenceBERT (SBERT) in the sentence vector evaluation using semantic textual similarity (STS) tasks and the SentEval toolkit.

Sentence transition matrix: An efficient approach that preserves sentence semantics

Sentence embedding is an influential research topic in natural language processing (NLP). Generation of sentence vectors that reflect the intrinsic meaning of sentences is crucial for improving performance in various NLP tasks. Therefore, numerous supervised and unsupervised sentence-representation approaches have been proposed since the advent of the distributed representation of words. These approaches have been evaluated on semantic textual similarity (STS) tasks designed to measure the degree of semantic information preservation; neural network-based supervised embedding models typically deliver state-of-the-art performance. However, these models have limitations in that they have numerous learnable parameters and thus require large amounts of specific types of labeled training data. Pretrained language model-based approaches, which have become a predominant trend in the NLP field, alleviate this issue to some extent; however, it is still necessary to collect sufficient labeled data for the fine-tuning process is still necessary. Herein, we propose an efficient approach that learns a transition matrix tuning a sentence embedding vector to capture the latent semantic meaning. Our proposed method has two practical advantages: (1) it can be applied to any sentence embedding method, and (2) it can deliver robust performance in STS tasks with only a few training examples.

Efficient natural language classification algorithm for detecting duplicate unsupervised features

This paper focuses on capturing the meaning of Natural Language Understanding (NLU) text features to detect the duplicate unsupervised features. The NLU features are compared with lexical approaches to prove the suitable classification technique. The transfer-learning approach is utilized to train the extraction of features on the Semantic Textual Similarity (STS) task. All features are evaluated with two types of datasets that belong to Bosch bug and Wikipedia article reports. This study aims to structure the recent research efforts by comparing NLU concepts for featuring semantics of text and applying it to IR. The main contribution of this paper is a comparative study of semantic similarity measurements. The experimental results demonstrate the Term Frequency–Inverse Document Frequency (TF-IDF) feature results on both datasets with reasonable vocabulary size. It indicates that the Bidirectional Long Short Term Memory (BiLSTM) can learn the structure of a sentence to improve the classification.

Sense representations for Portuguese: experiments with sense embeddings and deep neural language models

Sense representations have gone beyond word representations like Word2Vec, GloVe and FastText and achieved innovative performance on a wide range of natural language processing tasks. Although very useful in many applications, the traditional approaches for generating word embeddings have a strict drawback: they produce a single vector representation for a given word ignoring the fact that ambiguous words can assume different meanings. In this paper, we explore unsupervised sense representations which, different from traditional word embeddings, are able to induce different senses of a word by analyzing its contextual semantics in a text. The unsupervised sense representations investigated in this paper are: sense embeddings and deep neural language models. We present the first experiments carried out for generating sense embeddings for Portuguese. Our experiments show that the sense embedding model (Sense2vec) outperformed traditional word embeddings in syntactic and semantic analogies task, proving that the language resource generated here can improve the performance of NLP tasks in Portuguese. We also evaluated the performance of pre-trained deep neural language models (ELMo and BERT) in two transfer learning approaches: feature based and fine-tuning, in the semantic textual similarity task. Our experiments indicate that the fine tuned Multilingual and Portuguese BERT language models were able to achieve better accuracy than the ELMo model and baselines.

Exploiting Syntactic and Semantic Information for Textual Similarity Estimation

The textual similarity task, which measures the similarity between two text pieces, has recently received much attention in the natural language processing (NLP) domain. However, due to the vagueness and diversity of language expression, only considering semantic or syntactic features, respectively, may cause the loss of critical textual knowledge. This paper proposes a new type of structure tree for sentence representation, which exploits both syntactic (structural) and semantic information known as the weight vector dependency tree (WVD-tree). WVD-tree comprises structure trees with syntactic information along with word vectors representing semantic information of the sentences. Further, Gaussian attention weight is proposed for better capturing important semantic features of sentences. Meanwhile, we design an enhanced tree kernel to calculate the common parts between two structures for similarity judgment. Finally, WVD-tree is tested on widely used semantic textual similarity tasks. The experimental results prove that WVD-tree can effectively improve the accuracy of sentence similarity judgments.

The 2019 n2c2/OHNLP Track on Clinical Semantic Textual Similarity: Overview.

BackgroundSemantic textual similarity is a common task in the general English domain to assess the degree to which the underlying semantics of 2 text segments are equivalent to each other. Clinical Semantic Textual Similarity (ClinicalSTS) is the semantic textual similarity task in the clinical domain that attempts to measure the degree of semantic equivalence between 2 snippets of clinical text. Due to the frequent use of templates in the Electronic Health Record system, a large amount of redundant text exists in clinical notes, making ClinicalSTS crucial for the secondary use of clinical text in downstream clinical natural language processing applications, such as clinical text summarization, clinical semantics extraction, and clinical information retrieval.ObjectiveOur objective was to release ClinicalSTS data sets and to motivate natural language processing and biomedical informatics communities to tackle semantic text similarity tasks in the clinical domain.MethodsWe organized the first BioCreative/OHNLP ClinicalSTS shared task in 2018 by making available a real-world ClinicalSTS data set. We continued the shared task in 2019 in collaboration with National NLP Clinical Challenges (n2c2) and the Open Health Natural Language Processing (OHNLP) consortium and organized the 2019 n2c2/OHNLP ClinicalSTS track. We released a larger ClinicalSTS data set comprising 1642 clinical sentence pairs, including 1068 pairs from the 2018 shared task and 1006 new pairs from 2 electronic health record systems, GE and Epic. We released 80% (1642/2054) of the data to participating teams to develop and fine-tune the semantic textual similarity systems and used the remaining 20% (412/2054) as blind testing to evaluate their systems. The workshop was held in conjunction with the American Medical Informatics Association 2019 Annual Symposium.ResultsOf the 78 international teams that signed on to the n2c2/OHNLP ClinicalSTS shared task, 33 produced a total of 87 valid system submissions. The top 3 systems were generated by IBM Research, the National Center for Biotechnology Information, and the University of Florida, with Pearson correlations of r=.9010, r=.8967, and r=.8864, respectively. Most top-performing systems used state-of-the-art neural language models, such as BERT and XLNet, and state-of-the-art training schemas in deep learning, such as pretraining and fine-tuning schema, and multitask learning. Overall, the participating systems performed better on the Epic sentence pairs than on the GE sentence pairs, despite a much larger portion of the training data being GE sentence pairs.ConclusionsThe 2019 n2c2/OHNLP ClinicalSTS shared task focused on computing semantic similarity for clinical text sentences generated from clinical notes in the real world. It attracted a large number of international teams. The ClinicalSTS shared task could continue to serve as a venue for researchers in natural language processing and medical informatics communities to develop and improve semantic textual similarity techniques for clinical text.