Sentence Level Research Articles

A Helium Speech Correction Method Based on Generative Adversarial Networks

The distortion of helium speech caused by helium−oxygen gas mixtures significantly impacts the safety and communication efficiency of saturation divers. Although existing correction methods have shown some effectiveness in improving the intelligibility of helium speech, challenges remain in enhancing clarity and high−pitch correction. To address the issue of degraded speech quality post−correction, a novel helium speech correction method based on generative adversarial networks (GANs) is proposed. Firstly, a new helium speech dataset is introduced, which includes isolated words and continuous speech in both Chinese and English. By training and testing on both isolated words and continuous passages, the correction capability of the model can be accurately evaluated. Secondly, a new evaluation system for helium speech correction is proposed, which partially fills the gap in current helium speech evaluation metrics. This system uses comprehensive similarity to evaluate the similarity of keywords at the sentence level, thus assessing the correction results of helium speech from both word and sentence dimensions. Lastly, a GAN−based helium speech correction method is designed. This method solves the problems of pitch period distortion and formant shift in helium speech by introducing an adaptive speech segmentation algorithm and a fusion loss function and significantly improves the clarity and intelligibility of corrected helium speech. The experimental results show that the corrected helium speech is improved in clarity and intelligibility, which shows its practical value and application potential.

Influences of Attentional Focus on Across- and Within-Sentence Variability in Adults Who Do and Do Not Stutter.

Research has found an advantage to maintaining an external attentional focus while speaking as an increase in accuracy and a decrease in across-sentence variability has been found when producing oral-motor and speech tasks. What is not clear is how attention affects articulatory variability both across and within sentences, or how attention affects articulatory control in speakers who stutter. The purpose of this study was to investigate the effects of an internal versus external attention focus on articulatory variability at the sentence level. This study used linear (spatial-temporal index [STI]) and nonlinear (recurrence quantification analysis [RQA]) indices to measure lip aperture variability in 10 adults who stutter (AWS) and 15 adults who do not stutter (ANS) while they repeated sentences under an internal versus external attentional focus, virtual reality task (withVR.app; retrieved December 2023 from https://therapy.withvr.app). Four RQA measures were used to calculate within sentence variability including percent recurrence, percent determinism (%DET), stability (MAXLINE), and stationarity (TREND). Sentence duration measures were also obtained. AWS' movement durations were significantly longer than those of the ANS across conditions, and the AWS were more affected by the attentional focus shifts as their speech rate significantly increased when speaking with an external focus. AWS' speech patterns were also significantly more deterministic (%DET) and stable (MAXLINE) across attentional focus conditions compared to those of the ANS. Both groups showed an effect from attentional shifts as they exhibited less variability (i.e., more consistent) across sentences (STI) and less determinism (%DET) and stability (MAXLINE) within sentences when repeating sentences under an external attentional focus. STI values were not significantly different between the AWS and ANS for the internal or external attentional focus tasks. There were no significant main effects for group or condition for TREND; however, a main effect for sentence type was found. Results suggest that AWS use a more restrictive and less flexible approach to movement and that an external focus fosters more flexibility and thus responsiveness to external factors.

Academic language use in middle school informational writing.

Learning to write the complex academic language (AL) associated with a discipline (like science) is a critical task in education, with middle school being a key developmental period. However, we need more research to guide how we assess students' learning to write AL, especially if we want to create assessment that guides more effective instruction. We evaluated middle school students' informational writing for six different measures of AL to determine which ones were most strongly related to writing quality and were most indicative of the unique features of informational writing. We also examined which metrics were sensitive to growth across middle school. Our sample consists of informational compositions from 285 students in Grade 5 (n = 175) and Grade 8 (n = 110) in a Midwestern state in the United States. Path modelling was used to determine the degree to which the AL metrics are associated with writing quality and narrativity in each grade. Overall, the six measures of AL explained 70% of the variance in students' writing quality. We found that a new measure, number of long words, outperformed other more established measures at the word level and should be used in assessment of informational writing quality. We also found that automated scores at the sentence level better detected development across middle school grade levels than typical rubrics of writing quality. Results provide promising avenues for the assessment of malleable aspects of AL in middle school informational writing.

Research Trends in the Evaluation and Treatment of Speech Disorders Using Automatic Speech Recognition (ASR)

Purpose: The purpose of this study is to examine recent research trends regarding automatic speech recognition (ASR), which is used in the evaluation and intervention of speech disorders.Methods: Through a search engine, articles published in domestic journals were searched. A total of 27 papers were selected from the searched documents and analyzed according to the year, research subject, speech task, and ASR system.Results: The years with the most research was done in 2019~2021. The subjects who most frequently underwent speech evaluation and treatment using ASR system were those with dysarthria. The speech production tasks used to utilize ASR were at the word and sentence level, and commercialized and non-commercialized ASR systems were used similarly.Conclusion: These results might be used for the fundamental data to establish the most suitable evaluation methods and intervention plans for patients with speech disorders in clinical and research fields.

의사소통적 발음 교육을 위한 한국어 교재 분석 연구

Objectives The purpose of this paper is to analyze Korean textbooks for communicative pronunciation education. Methods Three types of textbooks developed by university-affiliated language education institutes were selected as analysis targets. As analysis tools, items that were synthesized and reconstructed from the textbook analysis frameworks of Seo (2001) and Lee (2001) were used. Results The result shows that the scope of teaching items was expanded, and it was necessary to secure areas equivalent to vocabulary and grammar, select teaching items for continuous and repetitive learning, and provide detailed grading. In addition, the educational effectiveness of the explicitness or implicitness of linguistic rule explanations had to be examined, and audiovisual materials should be abundantly utilized. The practice activities were planned to involve step-by-step practice from the word and sentence level to the discourse level, but the instructions were not clear, and even if there were directions, practice method was monotonous. For communicative pronunciation education, instructions and guidelines had to be specified and a variety of interactive and meaningful exercises had to be included. As for learning strategies, only simple repetition, memorization strategies, and strategies using visual images were induced. Effective pronunciation learning strategies should be found and learning strategies had to be developed in a variety of ways. Conclusions Considering practical application, a plan to revitalize pronunciation education was proposed while maintaining the current textbook. Improvement measures were discussed focusing on the practice stage. Integrated practice with other micro and macro skills and clear practice method instructions were emphasized. Practice and use were claimed at the word, sentence, and discourse levels in the order of presentation-controlled practice-meaningful practice-use. Interactive activities such as pair or group activities were considered.

Listening comprehension in a home language: a case of Russian in Germany.

Listening comprehension is central to language learning, yet it remains the least understood and least researched skill. This statement is still relevant today, as there is insufficient research to explore listening comprehension from the perspective of family-related multilingualism and to consider the complete linguistic repertoire of multilingual speakers. Moreover, with regard to home language, listening comprehension is assumed to be a more developed language competence than reading or writing. Based on the mentioned research, the aim of the present study is to investigate listening comprehension and its influencing factors specifically in German-Russian simultaneous bilinguals aged 13-19 (n = 99) by considering the home- and majority language. The study uses quantitative data collection methods such as linguistic tests in Russian and German for the elicitation in different levels of listening and questionnaires for strategy use and background. The research questions are as follows: What does the language proficiency and input in Russian look like? (1) Regarding listening comprehension in Russian as a home language, are there differences between the speakers within comprehension on different levels, e.g., is sound decoding easier than sentence parsing? (2) If there are differences in Russian as a home language, which linguistic and background variables can correlate with the performance of listening comprehension on its different levels? Concerning the first research question, the results show relevant differences between four different levels of listening comprehension (phoneme, word, sentence, and text level), which strengthened the assumed complexity of listening comprehension in the home language. In addition, the results show different connections between the listening comprehension competence and the input from different family members, as well as exposure to film and television in the home and majority language.

SỬ DỤNG GOOGLE DOCS VÀ MERCI APP TRONG LỚP HỌC VIẾT TIẾNG PHÁP

This study compares the effectiveness of two automated feedback tools, Google Documents, which will be further addressed as Google Docs, and Merci App, in improving writing ability in French as a foreign language. Forty second-year students used these two reviewer tools to correct their B2-level texts. The data includes 80 student drafts edited on two tools and a survey questionnaire after using these two applications. The results show that both tools help students correct many errors at the sentence, word, and fragment levels but cannot correct errors at the inter-sentence, idea, and coherence levels. Google Docs detects more grammatical errors than Merci App, but Merci App is superior when it comes to finding errors in spelling, vocabulary, expression, and punctuation. Students use the application's error correction suggestions to correct their papers at a very high rate, 92.6% with Google Docs and 81.7% with Merci App. The number of students who prefer Google Docs is also higher than the number of students who prefer Merci App. The study proposes to use both tools to correct students' writing errors and to help students become familiar with error explanations in Merci App to help them systematize language knowledge and develop auto-learning capacity.

Hearing, speech, and language outcomes in school-aged children after cleft palate repair.

Following primary cleft palate repair, individuals face a heightened risk of hearing problems, particularly conductive hearing loss, compensatory articulation disorders (CADs), resonance disorders, delayed speech and language development, and voice disorders. This study aimed to investigate the prevalence and impact of these challenges in children with cleft palate with or without cleft lip (CP± L). This cross-sectional study included 38 children with CP± L aged 5 to 13 years. A comprehensive evaluation involved audiological assessments (audiograms, tympanograms) by an audiologist and speech-language pathology assessments (Thai Speech Parameters for Patients with Cleft Palate, Articulation Screening Test) by speech-language pathologists. The prevalence of hearing loss affected 27.63% of participants (21 out of 76 ears) and majority of cases involved conductive hearing loss. Velar substitution was the most common CAD, followed by nasalized voiced pressure consonants, phoneme-specific nasal air emission, and pharyngeal substitution. A moderate correlation was found between these CAD patterns and hypernasality at the word, sentence, and screening levels (r= 0.44, p< 0.01; r= 0.43, p< 0.01; and r= 0.40, p= 0.01). For summary, the most common type of hearing loss was conductive hearing loss. The predominant CAD pattern was velar substitution. The protocol could be designed to enhance early improvement in hearing and articulation, thereby supporting academic achievement and long-term quality of life.

A polymorphic word in old Romanian (adjective, noun, adverb and interjection): ferice ʻgood/blessed/well, happyʼ

Consulting the main Romanian lexicographic works and taking into account some previous contributions on Romanian (diachronic) morphosyntax, the author aims to discuss the meanings of the word ferice ‘good/blessed/well, happy’ and its morphological values or syntactic implications, as they are presented in the old Romanian writings, more precisely in some old texts integrated into the SOR Corpus, enriched with new sources recently published. To this day, despite the antiquity of the word in question, it has been discussed only sequentially, which calls for revealing its morphosyntactic behavior. Consulting the main Romanian lexicographic works and taking into account some previous contributions on Romanian (diachronic) morphosyntax, the author aims to discuss the meanings of the word ferice ‘good/blessed/well, happy’ and its morphological values or syntactic implications, as they are presented in the old Romanian writings, more precisely in some old texts integrated into the SOR Corpus, enriched with new sources recently published. To this day, despite the antiquity of the word in question, it has been discussed only sequentially, which calls for revealing its morphosyntactic behavior. Therefore, it will be possible to establish much better the place and role of the word ferice ‘good/blessed/well, happy’ at the sentence level (in the old language), proposing some relevant analytical solutions, leading to some interpretative repositioning, so necessary for a proper understanding of language facts that are not always clear.

Cortical encoding of hierarchical linguistic information when syllabic rhythms are obscured by echoes

In speech perception, low-frequency cortical activity tracks hierarchical linguistic units (e.g., syllables, phrases, and sentences) on top of acoustic features (e.g., speech envelope). Since the fluctuation of speech envelope typically corresponds to the syllabic boundaries, one common interpretation is that the acoustic envelope underlies the extraction of discrete syllables from continuous speech for subsequent linguistic processing. However, it remains unclear whether and how cortical activity encodes linguistic information when the speech envelope does not provide acoustic correlates of syllables. To address the issue, we introduced a frequency-tagging speech stream where the syllabic rhythm was obscured by echoic envelopes and investigated neural encoding of hierarchical linguistic information using electroencephalography (EEG). When listeners attended to the echoic speech, cortical activity showed reliable tracking of syllable, phrase, and sentence levels, among which the higher-level linguistic units elicited more robust neural responses. When attention was diverted from the echoic speech, reliable neural tracking of the syllable level was also observed in contrast to deteriorated neural tracking of the phrase and sentence levels. Further analyses revealed that the envelope aligned with the syllabic rhythm could be recovered from the echoic speech through a neural adaptation model, and the reconstructed envelope yielded higher predictive power for the neural tracking responses than either the original echoic envelope or anechoic envelope. Taken together, these results suggest that neural adaptation and attentional modulation jointly contribute to neural encoding of linguistic information in distorted speech where the syllabic rhythm is obscured by echoes.

Cross-biased contrastive learning for answer selection with dual-tower structure

A large number of unanswered products-related questions appear in the E-commerce platforms, necessitating the deployment of question-answering models to automatically provide precise responses for the user. However, the substantial absence of ground truth presents challenges for implementing supervised learning, and existing unsupervised contrastive learning methods have not addressed this issue fundamentally, with limitations in integrating multi-level features. This paper presents a dual-tower cross-biased contrastive learning model for answer selection, named CCLM-AS. By taking questions and augmented product reviews as positive pairs for unsupervised contrastive learning and employing a dual-tower structure encoder, CCLM-AS learns sentence representations from unlabeled data effectively. Furthermore, the designed training objective realizes multi-level feature integration, including character, sentence and dialogue levels, which enhances the model's inference ability. The experimental results on AmazonQA dataset indicate that our proposed model outperforms existing contrastive learning-based sentence representation models and also achieves comparable performance to supervised answer selection models. This demonstrates that CCLM-AS can not only alleviate the data sparsity problem effectively but also retain the excellent performance.

Embodied cognition and L2 sentence comprehension: an eye-tracking study of motor representations.

Evidence from neuroscience and behavioral research has indicated that language meaning is grounded in our motor-perceptual experiences of the world. However, the question of whether motor embodiment occurs at the sentence level in L2 (second language) comprehension has been raised. Furthermore, existing studies on motor embodiment in L2 have primarily focused on the lexical and phrasal levels, often providing conflicting and indeterminate results. Therefore, to address this gap, the present eye-tracking study aimed to explore the embodied mental representations formed during the reading comprehension of L2 action sentences. Specifically, it sought to identify the types of motor representations formed during L2 action sentence comprehension and the extent to which these representations are motor embodied. A total of 56 advanced L2 learners participated in a Sentence-Picture Verification Task, during which their response times (RTs) and eye movements were recorded. Each sentence-picture pair depicted an action that either matched or mismatched the action implied by the sentence. Data analysis focused on areas of interest around the body effectors. RTs in the mismatch condition indicated an impeding effect. Furthermore, fixations on the body effector executing an action were longer in the mismatch condition, especially in late eye-movement measures.

Exploring Sources of Errors in L3 English Production of Natives of Moroccan Arabic

Language learners’ writing products can provide insights into how these learners acquire the target language through the analysis of the committed errors. Indeed, research has revealed that error analysis in writing essays is crucial for understanding and improving language acquisition by identifying common errors, and their sources, and informing teaching strategies to address these issues. The current paper aims to identify, describe and analyze errors committed by Moroccan secondary school students in writing. The study also attempts to determine some of the possible causes of the errors to offer an understanding of how linguistic interference influences the foreign language learning process. To this end, a sample of tasks written in English by a group of 20 secondary school students was collected. Then, an analysis procedure was conducted to analyze errors and explain their sources. Hence, this study adopts a mixed-method approach design to generate qualitative as well as quantitative data by identifying error categories and calculating the total number of error types and sources, respectively. The findings show that the errors are committed both at the word as well as the sentential level including lexical, grammatical, substance, and syntactic errors. Besides, the results indicate that learners commit errors in their English writings partly due to interference of Moroccan or Standard Arabic and French, but mostly due to other developmental sources within the scope of the target language. These findings offer valuable insights into how learners acquire a second or foreign language and what obstacles they encounter in the learning process and, thus, could suggest some implications to reduce the errors caused by negative language transfer and improve learners’ language proficiency.

Large scale structure-aware pronoun resolution using quantum natural language processing

Natural language consists of words, sentences, and larger units. Guided by grammatical structure, words compose with each other to form sentences. Similarly, guided by discourse structure, sentences compose with each other to form dialogues and documents. Classical machine learning algorithms have achieved significant success in learning the meanings of words. When it comes to meanings of sentences and discourse units, they however fall short of being compositional. The DisCoCat model of meaning, introduced by Clark, Coecke, and Sadrzadeh in 2010, provides a solution—at the sentence level—using higher-order tensors, the learning of which has been a challenge. A recent initiative known as Quantum Natural Language Processing (QNLP) introduces a translation between the DisCoCat tensors and Variational Quantum Circuits (VQC). This offers the potential of learning these higher-order tensors more efficiently when the circuits are executed on quantum computers. In previous work, we lifted the DisCoCat framework from the sentence level to the discourse level using a Fock space semantics. In this paper, we extend the DisCoCat-VQC translation to this semantics and experiment with it in a discourse task. We develop a massive dataset with 16,400 entries inspired by a major coreference resolution task, known as the Winograd Schema Challenge, proposed as a test of machine intelligence. Noisy and noiseless simulations were executed on IBMQ software, and the parameters of the discourse VQCs were learnt. The model converged to 77.5% accuracy, surpassing a Bag-of-Words model that neglects any structure. It also outperformed 2 out of 3 state-of-the-art classical coreference resolution architectures. These findings highlight the significant potential of quantum machine learning in advancing discourse analysis and structured natural language processing.

RegulaTome: a corpus of typed, directed, and signed relations between biomedical entities in the scientific literature.

In the field of biomedical text mining, the ability to extract relations from the literature is crucial for advancing both theoretical research and practical applications. There is a notable shortage of corpora designed to enhance the extraction of multiple types of relations, particularly focusing on proteins and protein-containing entities such as complexes and families, as well as chemicals. In this work, we present RegulaTome, a corpus that overcomes the limitations of several existing biomedical relation extraction (RE) corpora, many of which concentrate on single-type relations at the sentence level. RegulaTome stands out by offering 16 961 relations annotated in >2500 documents, making it the most extensive dataset of its kind to date. This corpus is specifically designed to cover a broader spectrum of >40 relation types beyond those traditionally explored, setting a new benchmark in the complexity and depth of biomedical RE tasks. Our corpus both broadens the scope of detected relations and allows for achieving noteworthy accuracy in RE. A transformer-based model trained on this corpus has demonstrated a promising F1-score (66.6%) for a task of this complexity, underscoring the effectiveness of our approach in accurately identifying and categorizing a wide array of biological relations. This achievement highlights RegulaTome's potential to significantly contribute to the development of more sophisticated, efficient, and accurate RE systems to tackle biomedical tasks. Finally, a run of the trained RE system on all PubMed abstracts and PMC Open Access full-text documents resulted in >18 million relations, extracted from the entire biomedical literature.

Determinants of Multilevel Discourse Outcomes in Anomia Treatment for Aphasia.

Individuals with aphasia identify discourse-level communication (i.e., language in use) as a high priority for treatment. The central premise of most aphasia treatments is that restoring language at the phoneme, word, and/or sentence level will generalize to discourse. However, treatment-related changes in discourse-level communication are modest, are poorly understood, and vary greatly among individuals with aphasia. In response, this study consisted of a multilevel discourse analysis of archival, monologic discourse outcomes across two high-intensity Semantic Feature Analysis (SFA) clinical trials. Aim 1 evaluated changes in theoretically motivated discourse outcomes representing lexical-semantic processing, lexical diversity, grammatical complexity, and discourse informativeness. Aim 2 explored the potential moderating role of nonlanguage cognitive factors (semantic memory, divided attention, and executive function) on discourse outcomes. This study was a retrospective analysis of archival monologic discourse outcomes after intensive SFA for n = 60 (Aim 1) and a subset n = 44 (Aim 2). Outcome measures included lexical-semantic processing (% semantic errors), lexical diversity (moving average type-token ratio), grammatical complexity (mean utterance length), and discourse informativeness (% correct information units). Bayesian generalized mixed-effects models were used to examine changes across four study time points: enrollment, entry, exit, and 1-month follow-up. The present study found no evidence for meaningful or statistically reliable improvements in monologue discourse performance after SFA when measured using standard, general-topic discourse stimuli. There was weak and inconsistent evidence that nonlanguage cognitive factors may play a role in moderating treatment response. These findings indicate a clear need to pair theoretically informed treatments designed to facilitate generalization to discourse with intentional measurement paradigms designed to capture it. Furthermore, there is a clear need to examine how established treatments, restorative or compensatory, can better facilitate generalization to discourse-level communication. These priorities are critical for meaningfully improving everyday communication and reducing the profound communication and psychosocial consequences of aphasia. https://doi.org/10.23641/asha.26524081.

Indonesian-English Textual Similarity Detection Using Universal Sentence Encoder (USE) and Facebook AI Similarity Search (FAISS)

The tremendous development in Natural Language Processing (NLP) has enabled the detection of bilingual and multilingual textual similarity. One of the main challenges of the Textual Similarity Detection (TSD) system lies in learning effective text representation. The research focuses on identifying similar texts between Indonesian and English across a broad range of semantic similarity spectrums. The primary challenge is generating English and Indonesian dense vector representation, a.k.a. embeddings that share a single vector space. Through trial and error, the research proposes using the Universal Sentence Encoder (USE) model to construct bilingual embeddings and FAISS to index the bilingual dataset. The comparison between query vectors and index vectors is done using two approaches: the heuristic comparison with Euclidian distance and a clustering algorithm, Approximate Nearest Neighbors (ANN). The system is tested with four different semantic granularities, two text granularities, and evaluation metrics with a cutoff value of k={2,10}. Four semantic granularities used are highly similar or near duplicate, Semantic Entailment (SE), Topically Related (TR), and Out of Topic (OOT), while the text granularities take on the sentence and paragraph levels. The experimental results demonstrate that the proposed system successfully ranks similar texts in different languages within the top ten. It has been proven by the highest F1@2 score of 0.96 for the near duplicate category on the sentence level. Unlike the near-duplicate category, the highest F1 scores of 0.77 and 0.89 are shown by the SE and TR categories, respectively. The experiment results also show a high correlation between text and semantic granularity.

Exploring the Effectiveness of Shallow and L2 Learner-Suitable Textual Features for Supervised and Unsupervised Sentence-Based Readability Assessment

Simplicity in information found online is in demand from diverse user groups seeking better text comprehension and consumption of information in an easy and timely manner. Readability assessment, particularly at the sentence level, plays a vital role in aiding specific demographics, such as language learners. In this paper, we research model evaluation metrics, strategies for model creation, and the predictive capacity of features and feature sets in assessing readability based on sentence complexity. Our primary objective is to classify sentences as either simple or complex, shifting the focus from entire paragraphs or texts to individual sentences. We approach this challenge as both a classification and clustering task. Additionally, we emphasize our tests on shallow features that, despite their simplistic nature and ease of use, seem to yield decent results. Leveraging the TextStat Python library and the WEKA toolkit, we employ a wide variety of shallow features and classifiers. By comparing the outcomes across different models, algorithms, and feature sets, we aim to offer valuable insights into optimizing the setup. We draw our data from sentences sourced from Wikipedia’s corpus, a widely accessed online encyclopedia catering to a broad audience. We strive to take a deeper look at what leads to greater readability classification in datasets that appeal to audiences such as Wikipedia’s, assisting in the development of improved models and new features for future applications with low feature extraction/processing times.

UNDERSTANDING LITERATURE VIA LINGUISTIC LENSES: THE EXPLORATION OF LANGUAGE APPROPRIATION STRATEGIES IN TARIQ RAHMAN'S MAMOON JI

The research paper explores different appropriation strategies of English language in Pakistani English literature. The research examined various appropriating techniques and the reasons behind using these techniques in short story Mamoon Ji by Tariq Rahman. English has become the lingua franca in the world. So, the writers tend to appropriate language in their writings from all over the world. The same phenomenon has been observed by the researchers in Tariq Rahman’s writing. The researchers analyze the text and apply Ashcroft’s theory of appropriation (2002) to investigate five appropriation strategies like glossing, untranslated words, syntactic fusion, vernacular transcription as well as code-switching. Many Urdu words have been analyzed along with Arabic words and Punjabi phrases at lexical and sentence levels. The focus remains on how Rahman employs the language appropriation techniques to convey themes, represents cultural contexts, and construct characters. The study provides the significant information in examining the issues, extracting desired information &amp; reaching outcomes. The outcome of current study has far-reaching implications from different viewpoints as it can be useful for pedagogy in teaching Pakistani English literature.

Lip reading using deep learning in Turkish language

&lt;p&gt;Computer vision is one of the most important areas of artificial intelligence and lip reading is one of the most important areas of computer vision. Lip-reading, which is more important in noisy environments or where there is no sound flow, is one of the working areas that can help the hearing-impaired people. There is no dataset in Turkish for lip reading, which there are different datasets at alphabet, word, and sentence level in different languages. The dataset of this study was created by the author and video data were collected from 72 people for 71 words. Audio streams were removed from the collected videos and a dataset was created using only images. Due to the small size of the dataset, the data was replicated with the Camtasia application. After the model of the research was designed and trained, the model was tested on adjectives, nouns, and verbs dataset and success rates of 71.8%, 71.88%, and 79.69% were obtained, respectively.&lt;/p&gt;