Lexical Semantic Similarity Research Articles

Flexible margins and multiple samples learning to enhance lexical semantic similarity

The advancement of deep learning and neural networks has led to the widespread adoption of neural word embeddings as a prominent lexical representation method in natural language processing. With the help of the neural language model trained by the contextual information of large scale text, the neural word embedding obtained by the neural language model captures more semantic correlation in the semantic space, while ignoring the semantic similarity. It will incur high computational cost and time costs during the training process of the model. To better inject semantic similarity into the distribution space and reduce time cost, we perform post processing learning of neural word embeddings using deep metric learning. This paper proposes a lexical enhancement method based on flexible margins and multiple samples learning. In this method, we embed the lexical entailment constraint relations into neural word embeddings. By categorizing the set of lexical constraints and penalizing the negative samples to different degrees according to the gap between categories, and allowing the positive and negative samples to learn from each other in the distributed space. The method we propose significantly improves neural word embeddings. By evaluating neural word embedded vocabulary similarity, the benchmark accuracy is improved to 75%. The method shows great competitiveness in text similarity tasks and text categorization tasks. These findings summarize research results and provide strong support for further applications.

Relational Analysis of College English Vocabulary - A Reflection Based on Semantic Association Network Modeling

Abstract Vocabulary is an essential and crucial part of college English learning. However, the lack of knowledge about the intrinsic connection between words has become a significant obstacle to vocabulary learning and teaching. In this study, we use a semantic association network to construct a relationship model between English words and introduce a tree kernel function into the syntactic analyzer to extract relationships between English words. The dictionary-based lexical semantic similarity calculation method is combined with a corpus-based English lexical semantic similarity calculation method for the extracted relations. Furthermore, the study creates datasets that relate to English vocabulary for simulation experiments. The results show that the more common sense phenomena there are, the higher the correlation value is. Furthermore, the similarity of university English vocabulary has a high degree of correlation with the type of specialized elements. The overall performance of the lexicon is ranked as synonym forest (0.431ρ/0.479r) > HowNet (0.249ρ/0.341r) > English WordNet (−0.051ρ/−0.003r), and the synonym forest method has the best performance, with the highest accuracy of 85.57% in classifying different lexical relations. The study offers valuable references and lessons for related research in natural language processing.

Feature extraction of search product based on multi-feature fusion-oriented to Chinese online reviews

The increasing Chinese online reviews contain rich product demand information, especially for search products. This study suggests a product feature extraction model from online reviews based on multi-feature fusion named PFEMF (products features extraction based on multi-feature fusion) model. Combining sentence and word characteristics of Chinese online reviews, the model explores the lexical features, frequency features, span features, and semantic similarity features of words. And then, they are fused to identify the features that customers are concerned about most by sequential relationship analysis. The identified product feature provides direction for product innovation and facilitates the product selection for customers. Finally, the study takes iPad Air as an example to prove this model. The results show that the extraction performance of the PFEMF model is superior to the traditional term frequency-inverse document frequency (tf-idf) algorithm, word span algorithm, and semantic similarity algorithm.

Canonical Sentence Processing and the Inferior Frontal Cortex: Is There a Connection?

The role of left inferior frontal cortex (LIFC) in canonical sentence comprehension is controversial. Many studies have found involvement of LIFC in sentence production or complex sentence comprehension, but negative or mixed results are often found in comprehension of simple or canonical sentences. We used voxel-, region-, and connectivity-based lesion symptom mapping (VLSM, RLSM, CLSM) in left-hemisphere chronic stroke survivors to investigate canonical sentence comprehension while controlling for lexical-semantic, executive, and phonological processes. We investigated how damage and disrupted white matter connectivity of LIFC and two other language-related regions, the left anterior temporal lobe (LATL) and posterior temporal-inferior parietal area (LpT-iP), affected sentence comprehension. VLSM and RLSM revealed that LIFC damage was not associated with canonical sentence comprehension measured by a sensibility judgment task. LIFC damage was associated instead with impairments in a lexical semantic similarity judgment task with high semantic/executive demands. Damage to the LpT-iP, specifically posterior middle temporal gyrus (pMTG), predicted worse sentence comprehension after controlling for visual lexical access, semantic knowledge, and auditory-verbal short-term memory (STM), but not auditory single-word comprehension, suggesting pMTG is vital for auditory language comprehension. CLSM revealed that disruption of left-lateralized white-matter connections from LIFC to LATL and LpT-iP was associated with worse sentence comprehension, controlling for performance in tasks related to lexical access, auditory word comprehension, and auditory-verbal STM. However, the LIFC connections were accounted for by the lexical semantic similarity judgment task, which had high semantic/executive demands. This suggests that LIFC connectivity is relevant to canonical sentence comprehension when task-related semantic/executive demands are high.

Bigram and Unigram Based Text Attack via Adaptive Monotonic Heuristic Search

Deep neural networks (DNNs) are known to be vulnerable to adversarial images, while their robustness in text classification are rarely studied. Several lines of text attack methods have been proposed in the literature, such as character-level, word-level, and sentence-level attacks. However, it is still a challenge to minimize the number of word distortions necessary to induce misclassification, while simultaneously ensuring the lexical correctness, syntactic correctness, and semantic similarity. In this paper, we propose the Bigram and Unigram based Monotonic Heuristic Search (BU-MHS) method to examine the vulnerability of deep models. Our method has three major merits. Firstly, we propose to attack text documents not only at the unigram word level but also at the bigram level to avoid producing meaningless outputs. Secondly, we propose a hybrid method to replace the input words with both their synonyms and sememe candidates, which greatly enriches potential substitutions compared to only using synonyms. Lastly, we design a search algorithm, i.e., Monotonic Heuristic Search (MHS), to determine the priority of word replacements, aiming to reduce the modification cost in an adversarial attack. We evaluate the effectiveness of BU-MHS on IMDB, AG's News, and Yahoo! Answers text datasets by attacking four state-of-the-art DNNs models. Experimental results show that our BU-MHS achieves the highest attack success rate by changing the smallest number of words compared with other existing models.

Semantic Data Set Construction from Human Clustering and Spatial Arrangement

Abstract Research into representation learning models of lexical semantics usually utilizes some form of intrinsic evaluation to ensure that the learned representations reflect human semantic judgments. Lexical semantic similarity estimation is a widely used evaluation method, but efforts have typically focused on pairwise judgments of words in isolation, or are limited to specific contexts and lexical stimuli. There are limitations with these approaches that either do not provide any context for judgments, and thereby ignore ambiguity, or provide very specific sentential contexts that cannot then be used to generate a larger lexical resource. Furthermore, similarity between more than two items is not considered. We provide a full description and analysis of our recently proposed methodology for large-scale data set construction that produces a semantic classification of a large sample of verbs in the first phase, as well as multi-way similarity judgments made within the resultant semantic classes in the second phase. The methodology uses a spatial multi-arrangement approach proposed in the field of cognitive neuroscience for capturing multi-way similarity judgments of visual stimuli. We have adapted this method to handle polysemous linguistic stimuli and much larger samples than previous work. We specifically target verbs, but the method can equally be applied to other parts of speech. We perform cluster analysis on the data from the first phase and demonstrate how this might be useful in the construction of a comprehensive verb resource. We also analyze the semantic information captured by the second phase and discuss the potential of the spatially induced similarity judgments to better reflect human notions of word similarity. We demonstrate how the resultant data set can be used for fine-grained analyses and evaluation of representation learning models on the intrinsic tasks of semantic clustering and semantic similarity. In particular, we find that stronger static word embedding methods still outperform lexical representations emerging from more recent pre-training methods, both on word-level similarity and clustering. Moreover, thanks to the data set’s vast coverage, we are able to compare the benefits of specializing vector representations for a particular type of external knowledge by evaluating FrameNet- and VerbNet-retrofitted models on specific semantic domains such as “Heat” or “Motion.”

Implementing WordNet Measures of Lexical Semantic Similarity in a Fuzzy Logic Programming System

AbstractThis paper introduces techniques to integrate WordNet into a Fuzzy Logic Programming system. Since WordNet relates words but does not give graded information on the relation between them, we have implemented standard similarity measures and new directives allowing the proximity equations linking two words to be generated with an approximation degree. Proximity equations are the key syntactic structures which, in addition to a weak unification algorithm, make a flexible query-answering process possible in this kind of programming language. This addition widens the scope of Fuzzy Logic Programming, allowing certain forms of lexical reasoning, and reinforcing Natural Language Processing (NLP) applications.

Story co-segmentation of Chinese broadcast news using weakly-supervised semantic similarity

This paper presents lexical story co-segmentation, a new approach to automatically extracting stories on the same topic from multiple Chinese broadcast news documents. Unlike topic tracking and detection, our approach needs not the guidance of well-trained topic models and can consistently segment the common stories from input documents. Following the MRF scheme, we construct a Gibbs energy function that feasibly balances the intra-doc and inter-doc lexical semantic dependencies and solve story co-segmentation as a binary labeling problem at sentence level. Due to the significance of measuring lexical semantic similarity in story co-segmentation, we propose a weakly-supervised correlated affinity graph (WSCAG) model to effectively derive the latent semantic similarities between Chinese words from the target corpus. Based on this, we are able to extend the classical cosine similarity by mapping the observed words distribution into the latent semantic space, which leads to a generalized lexical cosine similarity measurement. Extensive experiments on benchmark dataset validate the effectiveness of our story co-segmentation approach. Besides, we specifically demonstrate the superior performance of the proposed WSCAG semantic similarity measure over other state-of-the-art semantic measures in story co-segmentation.

Unsupervised measure of Chinese lexical semantic similarity using correlated graph model for news story segmentation

Abstract This paper presents a simple yet effective approach to unsupervisedly measuring Chinese lexical semantic similarity, and shows its promising performance in automatic story segmentation of Mandarin broadcast news. Our approach centers on the unsupervised correlated affinity graph (UCAG) model, which is initialized as a hybrid sparse graph, encoding both explicit word-to-word contextual correlations and latent word-to-character correlations within the given corpus. The UCAG model further diffuses the initial sparse correlations throughout the graph by parallel affinity propagation. This provides us with a dense, reliable, and corpus-specific lexical semantic similarity measure, which comes from purely unlabeled data. We then generalize the classical cosine similarity metric to effectively take soft similarities into account for story segmentation. Extensive experiments on benchmark datasets validate the superiority of the proposed similarity measure over previous measures. We specifically show that our similarity measure averagely helps to achieve 7.7% relative F1-score improvement to the accuracy of state-of-art normalized cuts (NCuts) based story segmentation on two holistic benchmark Mandarin broadcast news corpora, TDT2 and CCTV, and achieves 10.8% relative F1-score improvement on the detailed broadcast news subsets.

An Improved Tendentious Algorithm of Lexical Semantic

Lexical semantic tendentious recognition calculation is the base of the sentence, and the sentence tendentious recognition is the text and the chapter structure tendentious recognition foundation. Based on HowNet lexical semantic similarity calculation, according to the current vocabulary appraisable tendentious theory, the paper put forward an improved orientation algorithm. With experiment validation, in the same pair of benchmark words, accuracy rate has lager improve to above 90%. There are some values in practical application.

Priming by relational integration in perceptual identification and Stroop colour naming

Integrative priming is the facilitated recognition of a target word following a prime word with which it can be combined to produce a sub-type of the target (e.g., a lake bird is a type of bird). Such priming occurs even in the absence of lexical association, semantic similarity, or compound familiarity and so poses a challenge to current models of priming. The present research establishes integrative priming as a robust phenomenon across paradigms and tests whether it occurs controllably or uncontrollably. Target words (e.g., “bird”) were preceded by a prime word that was integratable (e.g., “lake”), associated and similar (e.g., “canary”), or unrelated (e.g., “trial”). Integrative priming was observed in a perceptual identification task that minimised strategic processing (Experiment 1) and in a Stroop colour naming task that penalised lexical integration (Experiment 2). Thus, like associative priming, integrative priming occurred uncontrollably. The results necessitate a distinct model of integrative priming, in which priming occurs automatically.

시맨틱 구문 트리 커널을 이용한 생명공학 분야 전문용어간 관계 식별 및 분류 연구

본 논문에서는 단백질 간 상호작용 자동 추출을 위해서 기존에 연구되어 높은 성능을 나타낸 구문 트리 커널을 확장한 시맨틱 구문 트리 커널을 제안한다. 기존 구문 트리 커널의 문제점은 구문 트리의 단말 노드를 구성하는 개별 어휘에 대한 단순 외형적 비교로 인해, 실제 의미적으로는 유사한 두 구문 트리의 커널 값이 상대적으로 낮아지는 현상이며 결국 상호작용 자동 추출의 전체 성능에 악영향을 줄 수 있다는 점이다. 본 논문에서는 두 구문 트리의 구문적 유사도(syntactic similarity)와 어휘 의미적 유사도(lexical semantic similarity)를 동시에 효과적으로 계산하여 이를 결합하는 새로운 커널을 고안하였다. 어휘 의미적 유사도 계산을 위해서 문맥 및 워드넷 기반의 어휘 중의성 해소 시스템과 이 시스템의 출력으로 도출되는 어휘 개념(WordNet synset)의 추상화를 통한 기존 커널의 확장을 시도하였다. 실험에서는 단백질 간 상호작용 추출(PPII, PPIC) 성능의 심층적 최적화를 위해서 기존의 SVM에서 지원되던 정규화 매개변수 외에 구문 트리 커널의 소멸인자와 시맨틱 구문 트리 커널의 어휘 추상화 인자를 새롭게 도입하였다. 이를 통해 구문 트리 커널을 적용함에 있어서 소멸인자 역할의 중요성을 확인할 수 있었고, 시맨틱 구문 트리 커널이 기존 시스템의 성능향상에 도움을 줄 수 있음을 실험적으로 보여주었다. 특히 단백질 간 상호작용 식별 문제보다도 비교적 난이도가 높은 상호작용 분류에 더욱 효과적임을 알 수 있었다.

A semantic analysis of grammatical class impairments: semantic representations of object nouns, action nouns and action verbs

The grammatical class distinction between nouns and verbs is largely parallel to the semantic distinction between objects and actions; in this paper, we explore the extent to which grammatical class effects (noun- or verb-specific naming deficits) can be explained by lexical semantic factors alone. In order to do so, we investigate lexical–semantic clustering properties, not only of nouns depicting objects and verbs depicting actions, but also of nouns depicting actions, which should exhibit some patterns of similarity to other nouns if grammatical class emerges on the basis of lexical semantics. We collected speaker-generated features and used self-organizing maps to model lexical–semantic similarity among words. We simulated lesions on the resulting map, finding patterns of object-noun/action-verb naming impairments consistent with those reported in the literature. Importantly, we found that action-nouns exhibited no tendency to be more similar to object-nouns than their corresponding action-verbs, a finding inconsistent with a semantic account of grammatical class.