Word Segmentation Research Articles

Efficient word segmentation is preserved in older adult readers: Evidence from eye movements during Chinese reading.

College-aged readers use efficient strategies to segment and recognize words in naturally unspaced Chinese text. Whether this capability changes across the adult lifespan is unknown, although segmenting words in unspaced text may be challenging for older readers due to visual and cognitive declines in older age, including poorer parafoveal processing of upcoming characters. Accordingly, we conducted two eye movement experiments to test for age differences in word segmentation, each with 48 young (18-30 years) and 36 older (65+ years) native Chinese readers. Following Zhou and Li (2021), we focused on the processing of "incremental" three-character words, like (meaning "kindergartens"), which contain an embedded two-character word (e.g., , meaning "children"). In Experiment 1, either the three-character word or its embedded word was presented as the target word in sentence contexts where the three-character word always was plausible, and the embedded word was either plausible or implausible. Both age groups produced similar plausibility effects, suggesting age constancy in accessing the embedded word early during ambiguity processing before ultimately assigning an incremental word analysis. Experiment 2 provided further evidence that both younger and older readers access the embedded word early during ambiguity processing, but rapidly select the appropriate (incremental) word. Crucially, the findings suggest that word segmentation strategies do not differ with age. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

What Do Self-Supervised Speech Models Know About Words?

Abstract Many self-supervised speech models (S3Ms) have been introduced over the last few years, improving performance and data efficiency on various speech tasks. However, these empirical successes alone do not give a complete picture of what is learned during pre-training. Recent work has begun analyzing how S3Ms encode certain properties, such as phonetic and speaker information, but we still lack a proper understanding of knowledge encoded at the word level and beyond. In this work, we use lightweight analysis methods to study segment-level linguistic properties—word identity, boundaries, pronunciation, syntactic features, and semantic features—encoded in S3Ms. We present a comparative study of layer-wise representations from ten S3Ms and find that (i) the frame-level representations within each word segment are not all equally informative, and (ii) the pre-training objective and model size heavily influence the accessibility and distribution of linguistic information across layers. We also find that on several tasks—word discrimination, word segmentation, and semantic sentence similarity—S3Ms trained with visual grounding outperform their speech-only counterparts. Finally, our task-based analyses demonstrate improved performance on word segmentation and acoustic word discrimination while using simpler methods than prior work.1

English Vocabulary Learning Aid System Using Digital Twin Wasserstein Generative Adversarial Network Optimized With Jelly Fish Optimization Algorithm

ABSTRACT Natural Language Processing (NLP) is a technology that permits computers to recognize human languages. Words are the fundamental unit of analysis in deep-level grammatical and semantic analysis. The main goal of NLP is typically word segmentation. Since the machine learning techniques cannot be directly applied to the practical issue of significant structural disparities between various data modalities in a multi-modal context. In this paper, English Vocabulary Learning Aid System Using Digital Twin Wasserstein Generative Adversarial Network Optimized using Jelly Fish Optimization Algorithm is proposed. The problematic of multiple modal data heterogeneity is handled by the feature extraction of Parameterized Local Maximum Synchro squeezing Transform and extract the features such as Phonetic features, sentence length, word embedding’s, part of speech tags, word frequencies, N-grams. Then, the Digital twin Wasserstein generative adversarial network classifies the English vocabulary to easy words, intermediate words, and difficult words. The performance of the proposed EVLS-DtwinWGAN-NLP approach attains 3.101%, 7.12%, 7.73% higher accuracy, 24.13%, 13.04%, 29.51% lower computation Time and 2.292%, 5.365%, 1.551% higher AUC compared with existing methods like Feature extraction and analysis of natural language processing for deep learning English language (EVLS-BiLSTM-NLP), State of art for semantic analysis of natural language processing (EVLS-SA-NLP) respectively.

Sentiment analysis of the awareness of environmental sustainability

This study examines the sentiment analysis of awareness of environmental sustainability. Environmental sustainability is the responsible management and utilization of Earth's natural resources to meet the needs of the present generation and ensure that future generations will access those resources. The awareness of environmental sustainability has been growing globally as people, businesses, and governments recognize the importance of preserving the planet for current and future generations. Sentiment analysis of environmental sustainability involves evaluating opinions, attitudes, and emotions expressed in texts related to environmental sustainability, and analyzing sentiment can provide insights into public perception, awareness, and engagement with environmental issues. This exploratory study's primary goal is to conduct social media opinion mining in the context of Thai people's environmental sustainability. The paper presented how to build a model of sentiment analysis with linguistic analysis, including data preprocessing steps, feature extraction, and model constructions. The techniques used in this research include Logistic Regression, Random Forests, Support Vector Machine, Word Segmentation and Bag of Words. The result shows that the model is able to categorize sentiment analysis opinions in the sustainability context primarily in positive terms. The positive sentiments suggest a sustained, long-term shift in awareness, or they might be influenced by specific events or trends. However, positive sentiment analysis results are expressed towards environmental sustainability initiatives, such as renewable energy projects, waste reduction efforts, or conservation programs. Moreover, public awareness plays a crucial role in influencing individual behavior, corporate practices, and government policies towards a more sustainable and environmentally conscious future.

Comparison and analysis of two methods for improving the accuracy of OpenNMT in literary and vernacular Chinese translation

As machine translation advances, challenges persist in achieving high accuracy when translating between Literary and Vernacular Chinese. Literary Chinese is known for its concise writing style, often characterized by monosyllabic words. In contrast, Vernacular Chinese incorporates more polysyllabic words. This study utilizes the OpenNMT system to address these challenges and employs different Classical Chinese word segmentation tools to train 2-layer Long Short-Term Memory and Transformer models. These models are then compared and analyzed to measure the improvement in precision. The research findings reveal that adopting a character-level-based word segmentation method for Classical Chinese, coupled with training the Transformer model using OpenNMT, significantly enhances precision. This outcome validates the current observation that existing Classical Chinese word segmentation methods lack sufficient accuracy, consequently impacting the quality of translations between Literary and Vernacular Chinese. By exploring and investigating these approaches, this study contributes to advancing machine translation techniques for improving accuracy in rendering Literary and Vernacular Chinese translations.

Comparing Differences in Jogging Support across Various Land Use Types in Urban Built-Up Areas Using User-Recommended Routes

Land use types other than specialized athletic fields provide a variety of jogging environments, addressing the shortage of urban fitness facilities and promoting urban health as well as sustainability. Currently, there is limited research comparing the differences in jogging support among various land use types, which can assist decision-makers in setting priorities and targeted strategies for urban renewal, especially in urban built-up areas with limited land resources. Initially, spatial information, statistical data, and recommendation reason text were extracted from recommended routes in mobile fitness apps and categorized into six land use types. Subsequently, spatial potential was measured through descriptive statistics, buffer area analysis, spatial autocorrelation analysis, and line density analysis. Environmental preferences were gauged by text analysis using jieba word segmentation and grouped word frequency calculation. Finally, the measurement results of different land uses were compared, including scale differences, spatial differentiation, environmental perception, and environmental elements. The research found that streets, residential areas, campuses, parks, and greenways possess significant potential to support jogging, particularly streets. These types of land use exhibit varying spatial potentials and attractions in environmental preferences. Targeted recommendations have been proposed to support the renewal of urban built-up areas and research in related fields.

Habilidade na escrita e nível de letramento de escreventes do Rio Grande do Sul nos séculos séculos XIX e XX

Studies such as those by Marquilhas (2000), Barbosa (2005, 2017), Santiago (2012), among others, propose that it is possible to characterize the ability of a writer, that is, his/her greater or lesser familiarity with writing, based on the observation of physical-graphical and graphic and phonetic aspects. This kind of description allows us to identify the level of literacy of the writers, which can help to compose their socio-linguistic profiles, especially when documents produced by non-distinguished people are contemplated. In this paper, we analyze five manuscripts of four writers from the cities of Santa Maria and Silveira Martins, both located in the state of Rio Grande do Sul (Southern Brazil), written in the late nineteenth century (1890) and early twentieth century (1901). These texts belong to the collection named Junta Intendencial of the Municipal Historic Archives of Santa Maria. The documents were photographed and transcribed faithfully in order to retain their original features. Issues related to the layout of the letters, the use of abbreviations, the substitution of segments (consonants and vowels), unconventional word segmentation, and the use of Latinized words were observed in the manuscripts so as to determine the closeness of the writers to the written code. In general, we found that all the writers have a low level of literacy.

A relational extraction approach based on multiple embedding representations and multi-head self-attention

Currently, word segmentation errors and polysemy problems are common in the field of Chinese relationship extraction. Although character-based model input can avoid word segmentation errors, in order to obtain the word information of a sentence, it is often necessary to introduce a dictionary or an external knowledge base to expand the word information, which requires a lot of manpower and time. In response to the above existing problems, this article uses characters as input, uses multiple embedding models to jointly form a character vector sequence, and obtains features containing character information through BiLSTM and attention layers; considering that convolutional neural networks are good at extracting local features, obtain features containing word information through multi-kernel convolutional layers and multi-head self-attention layers, and finally use a gating mechanism to fuse the features. The model was tested on the public SanWen data set and our own cultural-travel data set, and obtained F1 values of 61.22% and 60.26% respectively. Experimental results show that our method can achieve better relationship extraction effects without using word segmentation tools and without building a dictionary or external knowledge base, and the effect is better than most commonly used models currently.

Automatic analysis of public health service text based on character level convolutional neural network

The text information processing technology of public health service is one of the hot research topics at present. To improve the defects of public health service texts, such as inaccurate word segmentation, spelling errors and professional vocabulary understanding, this study designed a character-level deep neural network model on the characteristics of public health service texts. In this model, the bidirectional short and short time memory and the attention pooling operation layer are introduced to make the model better classify the text according to the context. In addition, counter perturbation is introduced in this study to improve the robustness and generalization ability of the model, thus improving its classification effect. The performance verification results show that the proposed model has better classification performance on the public health service text data set. The anti-disturbance samples generated by the model are all in the range of 0–0.2 when WMD deviation degree is measured, while most of the other methods are in the range of 0.4–0.6. The experimental object of this study is ultrasonic examination data. The experimental results show that the automatic analysis model of public health service text based on character level convolutional neural network constructed in this study has excellent accuracy and convergence speed, and has excellent performance in the classification of public health service text in different subject areas.

Sentiment analysis of online licensing service quality in the energy and mineral resources sector of the Republic of Indonesia

The Ministry of Energy and Mineral Resources of the Republic of Indonesia regularly assessed public satisfaction with its online licensing services. User rated their satisfaction at 3.42 on a scale of 4, below the organization's average of 3.53. Evaluating public service performance is crucial for quality improvement. Previous research relied solely on survey data to assess public satisfaction. This study goes further by analyzing user feedback in text form from an online licensing application to identify negative aspects of the service that need enhancement. The dataset spanned September 2019 to February 2023, with 24,112 entries. The choice of classification methods on the highest accuracy values among decision tree, random forest, naive bayes, stochastic gradient descent, logistic regression (LR), and k-nearest neighbor. The text data was converted into numerical form using CountVectorizer and term frequency-inverse document frequency (TF-IDF) techniques, along with unigrams and bigrams for dividing sentences into word segments. LR bigram CountVectorizer ranked highest with 89% for average precision, F1-score, and recall, compared to the other five classification methods. The sentiment analysis polarity level was 36.2% negative. Negative sentiment revealed expectations from the public to the ministry to improve the top three aspects: system, mechanism, and procedure; infrastructure and facilities; and service specification product types.

Non-native clear speech increases intelligibility but not through improved word segmentation: Evidence from a visual-world eye-tracking study

Listener-directed hyperarticulated clear speech produced by native (L1) talkers improves word segmentation and reduces lexical competition. Less is known about whether non-native (L2) clear speech also confers such benefit. In a visual-world eye-tracking study, we investigated if L2 clear speech improves word segmentation and the time course of the benefit for native listeners. Forty L1 English participants heard sentences produced in conversational and clear styles by a highly intelligible L2 English / L1 Spanish speaker with a discernable non-native accent. Sentences contained a target word (e.g., doll) with which a corresponding competitor overlapped phonemically (e.g., dolphin), creating temporary ambiguity with the target and the following word’s onset (e.g., doll found). Each recording was presented in quiet alongside pictures of the target, competitor, and two distractors. Participants were instructed to select the picture mentioned in the sentence they heard. No significant clear speech segmentation advantage was found; the proportion of looks to targets over competitors indicates similar time course of disambiguation in both conversational and clear speech. The results suggest that L2-accented clear speech with its deviations from the target-language- specific modifications and greater phonetic variability increases signal uncertainty resulting in no benefit for word segmentation even though word recognition was improved.

Sentiment analysis of online licensing service quality in the energy and mineral resources sector of the Republic of Indonesia

The Ministry of Energy and Mineral Resources of the Republic of Indonesia regularly assessed public satisfaction with its online licensing services. User rated their satisfaction at 3.42 on a scale of 4, below the organization's average of 3.53. Evaluating public service performance is crucial for quality improvement. Previous research relied solely on survey data to assess public satisfaction. This study goes further by analyzing user feedback in text form from an online licensing application to identify negative aspects of the service that need enhancement. The dataset spanned September 2019 to February 2023, with 24,112 entries. The choice of classification methods on the highest accuracy values among decision tree, random forest, naive bayes, stochastic gradient descent, logistic regression (LR), and k-nearest neighbor. The text data was converted into numerical form using CountVectorizer and term frequency-inverse document frequency (TF-IDF) techniques, along with unigrams and bigrams for dividing sentences into word segments. LR bigram CountVectorizer ranked highest with 89% for average precision, F1-score, and recall, compared to the other five classification methods. The sentiment analysis polarity level was 36.2% negative. Negative sentiment revealed expectations from the public to the ministry to improve the top three aspects: system, mechanism, and procedure; infrastructure and facilities; and service specification product types.

Computational Modeling of the Segmentation of Sentence Stimuli From an Infant Word-Finding Study.

Computational models of infant word-finding typically operate over transcriptions of infant-directed speech corpora. It is now possible to test models of word segmentation on speech materials, rather than transcriptions of speech. We propose that such modeling efforts be conducted over the speech of the experimental stimuli used in studies measuring infants' capacity for learning from spoken sentences. Correspondence with infant outcomes in such experiments is an appropriate benchmark for models of infants. We demonstrate such an analysis by applying the DP-Parser model of Algayres and colleagues to auditory stimuli used in infant psycholinguistic experiments by Pelucchi and colleagues. The DP-Parser model takes speech as input, and creates multiple overlapping embeddings from each utterance. Prospective words are identified as clusters of similar embedded segments. This allows segmentation of each utterance into possible words, using a dynamic programming method that maximizes the frequency of constituent segments. We show that DP-Parse mimics American English learners' performance in extracting words from Italian sentences, favoring the segmentation of words with high syllabic transitional probability. This kind of computational analysis over actual stimuli from infant experiments may be helpful in tuning future models to match human performance.

Can Infants Retain Statistically Segmented Words and Mappings Across a Delay?

Infants are sensitive to statistics in spoken language that aid word-form segmentation and immediate mapping to referents. However, it is not clear whether this sensitivity influences the formation and retention of word-referent mappings across a delay, two real-world challenges that learners must overcome. We tested how the timing of referent training, relative to familiarization with transitional probabilities (TPs) in speech, impacts English-learning 23-month-olds' ability to form and retain word-referent mappings. In Experiment 1, we tested infants' ability to retain TP information across a 10-min delay and use it in the service of word learning. Infants successfully mapped high-TP but not low-TP words to referents. In Experiment 2, infants readily mapped the same words even when they were unfamiliar. In Experiment 3, high- and low-TP word-referent mappings were trained immediately after familiarization, and infants readily remembered these associations 10 min later. In sum, although 23-month-old infants do not need strong statistics to map word forms to referents immediately, or to remember those mappings across a delay, infants are nevertheless sensitive to these statistics in the speech stream, and they influence mapping after a delay. These findings suggest that, by 23 months of age, sensitivity to statistics in speech may impact infants' language development by leading word forms with low coherence to be poorly mapped following even a short period of consolidation.

Validation of a Thai artificial chatmate designed for cheering up the elderly during the COVID-19 pandemic.

The COVID-19 pandemic severely affected populations of all age groups. The elderly are a high-risk group and are highly vulnerable to COVID-19. Assistive software chatbots can enhance the mental health status of the elderly by providing support and companionship. The objective of this study was to validate a Thai artificial chatmate for the elderly during the COVID-19 pandemic and floods. Chatbot design includes the establishment of a dataset and emotional word vectors in which data consisting of emotional sentences were converted into the word vector form using a pre-trained word2vec model. A word vector was then input into a convolutional neural network (CNN) and trained until the model converges using sentence embedding and similarity word segmentation. Sentence vectors were generated by averaging each word vector using an averaged vector method. For approximate similarity matching, the Annoy library was used to create the indices in tree sorting. Data were collected from 22 elderly and assessed by the Post-Study System Usability Questionnaire (PSSUQ). The study revealed that 72.73% of the respondents found the chatbot easy to learn and use, 63.64% of the respondents found the chatbot can autonomously determine the next course of action, and 59.09% of the respondents believed that troubleshooting guidelines were provided for overcoming errors. The accuracy of the chatbot providing a reasonable response is 56.20±13.99%. Most users were satisfied with the chatbot system. The proposed chatbot provided considerable essential insights into the development of assistance systems for the elderly during the coronavirus pandemic (COVID-19) and during the period of national disasters. The model can be expanded to other applications in the future.

Discrimination of Degrees of Foreign Accent across Different Speakers

Second-language learners often encounter communication challenges due to a foreign accent (FA) in their speech, influenced by their native language (L1). This FA can affect rhythm, intonation, stress, and the segmental domain, which consists of individual language sounds. This study looks into the segmental FA aspect, exploring listeners’ perceptions when Spanish interacts with English. Utilizing the SIAEW corpus, which replaces segments of English words with anticipated Spanish-accented realizations, we assess the ability of non-native listeners to discriminate degrees of accent across male and female voices. This research aims to determine the impact of voice consistency on detecting accentedness variations, studying participants from Japanese and Spanish. Results show that, while listeners are generally able to discriminate degrees of foreign accent across speakers, some segmental transformations convey a more clear distinction depending on the phonological representations of the native and accented realisations on the listener’s system. Another finding is that listeners tend to better discriminate degrees of accent when words are more native-like sounding.

Online biomedical named entities recognition by data and knowledge-driven model

Named entity recognition (NER) is an important task for the natural language processing of biomedical text. Currently, most NER studies standardized biomedical text, but NER for unstandardized biomedical text draws less attention from researchers. Named entities in online biomedical text exist with errors and polymorphisms, which negatively impact NER models’ performance and impede support from knowledge representation methods. In this paper, we propose a neural network method that can effectively recognize entities in unstandardized online medical/health text. We introduce a new pre-training scheme that uses large-scale online question-answering pairs to enhance transformers’ model capacity on online biomedical text. Moreover, we supply models with knowledge representations from a knowledge base called multi-channel knowledge labels, and this method overcomes the restriction from languages, like Chinese, that require word segmentation tools to represent knowledge. Our model outperforms other baseline methods significantly in experiments on a dataset for Chinese online medical entity recognition and achieves state-of-the-art results.

Phonological syllables allow children with developmental dyslexia to access words.

Learning to read is a middle-distance race for children worldwide. Most of them succeed in this acquisition with "normal" difficulties that ensue from the progressive (re)structuring of the phonological and orthographic systems. Evidence accumulated on reading difficulties in children with developmental dyslexia (DYS children, henceforth) shows a pervasive phonological deficit. However, the phonological deficit may not be due to degraded phonological representations but rather due to impaired access to them. This study focused on how and to what extent phonological syllables, which are essential reading units in French, were accessible to DYS children to segment and access words. We tested the assumption that DYS children did not strictly have pervasive degraded phonological representations but also have impaired access to phonological and orthographic representations. We administered a visually adapted word-spotting paradigm, engaging both sublexical processing and lexical access, with French native-speaking DYS children (N = 25; Mage in months = 121.6, SD = 3.0) compared with chronological age-matched peers (N = 25; Mage in months = 121.8, SD = 2.7; CA peers henceforth) and reading level-matched peers (N = 25; Mage in months = 94.0, SD = 4.6; RL peers henceforth). Although DYS children were slower and less accurate than CA and RL peers, we found that they used phonological syllables to access and segment words. However, they exhibit neither the classical inhibitory syllable frequency effect nor the lexical frequency effect, which is generally observed in typically developing children. Surprisingly, DYS children did not show strictly degraded phonological representations because they demonstrated phonological syllable-based segmentation abilities, particularly with high-frequency syllables. Their difficulties are rather interpreted in terms of impaired access to orthographic and phonological representations, which could be a direct effect of difficulties in generalizing and consolidating low-frequency syllables. We discuss these results regarding reading acquisition and the specificities of the French linguistic system.

The influence of language-specific properties on the role of consonants and vowels in a statistical learning task of an artificial language: A cross-linguistic comparison.

The contribution of consonants and vowels in spoken word processing has been widely investigated, and studies have found a phenomenon of a Consonantal bias (C-bias), indicating that consonants carry more weight than vowels. However, across languages, various patterns have been documented, including that of no preference or a reverse pattern of Vowel bias. A central question is how the manifestation of the C-bias is modulated by language-specific factors. This question can be addressed by cross-linguistic studies. Comparing native Hebrew and native English speakers, this study examines the relative importance of transitional probabilities between non-adjacent consonants as opposed to vowels during auditory statistical learning (SL) of an artificial language. Hebrew is interesting because its complex Semitic morphological structure has been found to play a central role in lexical access, allowing us to examine whether morphological properties can modulate the C-bias in early phases of speech perception, namely, word segmentation. As predicted, we found a significant interaction between language and consonant/vowel manipulation, with a higher performance in the consonantal condition than in the vowel condition for Hebrew speakers, namely, C-bias, and no consonant/vowel asymmetry among English speakers. We suggest that the observed interaction is morphologically anchored, indicating that phonological and morphological processes interact during early phases of auditory word perception.

Two-Step Word Segmentation Without Prior Knowledge of a Small Amount of Text

Two-Step Word Segmentation Without Prior Knowledge of a Small Amount of Text