PLOS ONE: [sortOrder=DATE_NEWEST_FIRST, sort=Date, newest first, q=subject:"Linguistics"]PLOShttps://journals.plos.org/plosone/webmaster@plos.orgaccelerating the publication of peer-reviewed sciencehttps://journals.plos.org/plosone/search/feed/atom?sortOrder=DATE_NEWEST_FIRST&unformattedQuery=subject:%22Linguistics%22&sort=Date,+newest+firstAll PLOS articles are Open Access.https://journals.plos.org/plosone/resource/img/favicon.icohttps://journals.plos.org/plosone/resource/img/favicon.ico2024-03-28T12:16:47ZAnomalous diffusion analysis of semantic evolution in major Indo-European languagesBogdán AsztalosGergely PallaDániel Czégel10.1371/journal.pone.02986502024-03-26T14:00:00Z2024-03-26T14:00:00Z<p>by Bogdán Asztalos, Gergely Palla, Dániel Czégel</p>
How do words change their meaning? Although semantic evolution is driven by a variety of distinct factors, including linguistic, societal, and technological ones, we find that there is one law that holds universally across five major Indo-European languages: that semantic evolution is subdiffusive. Using an automated pipeline of diachronic distributional semantic embedding that controls for underlying symmetries, we show that words follow stochastic trajectories in meaning space with an anomalous diffusion exponent <i>α</i> = 0.45 ± 0.05 across languages, in contrast with diffusing particles that follow <i>α</i> = 1. Randomization methods indicate that preserving temporal correlations in semantic change <i>directions</i> is necessary to recover strongly subdiffusive behavior; however, correlations in change <i>sizes</i> play an important role too. We furthermore show that strong subdiffusion is a robust phenomenon under a wide variety of choices in data analysis and interpretation, such as the choice of fitting an ensemble average of displacements or averaging best-fit exponents of individual word trajectories.Anomaly based multi-stage attack detection methodWei MaYunyun HouMingyu JinPengpeng Jian10.1371/journal.pone.03008212024-03-25T14:00:00Z2024-03-25T14:00:00Z<p>by Wei Ma, Yunyun Hou, Mingyu Jin, Pengpeng Jian</p>
Multi-stage attacks are one of the most critical security threats in the current cyberspace. To accurately identify multi-stage attacks, this paper proposes an anomaly-based multi-stage attack detection method. It constructs a Multi-Stage Profile (MSP) by modeling the stable system’s normal state to detect attack behaviors. Initially, the method employs Doc2Vec to vectorize alert messages generated by the intrusion detection systems (IDS), extracting profound inter-message correlations. Subsequently, Hidden Markov Models (HMM) are employed to model the normal system state, constructing an MSP, with relevant HMM parameters dynamically acquired via clustering algorithms. Finally, the detection of attacks is achieved by determining the anomaly threshold through the generation probability (GP). To evaluate the performance of the proposed method, experiments were conducted using three public datasets and compared with three advanced multi-stage attack detection methods. The experimental results demonstrate that our method achieves an accuracy of over 99% and precision of 100% in multi-stage attack detection. This confirms the effectiveness of our method in adapting to different attack scenarios and ultimately completing attack detection.Contribution of statistical learning in learning to read across languagesJinglei RenMin Wang10.1371/journal.pone.02986702024-03-25T14:00:00Z2024-03-25T14:00:00Z<p>by Jinglei Ren, Min Wang</p>
Statistical Learning (SL) refers to human’s ability to detect regularities from environment Kirkham, N. Z. (2002) & Saffran, J. R. (1996). There has been a growing interest in understanding how sensitivity to statistical regularities influences learning to read. The current study systematically examined whether and how non-linguistic SL, Chinese SL, and English SL contribute to Chinese and English word reading among native Chinese-speaking 4th, 6th and 8th graders who learn English as a second language (L2). Children showed above-chance learning across all SL tasks and across all grades. In addition, developmental improvements were shown across at least two of the three grade ranges on all SL tasks. In terms of the contribution of SL to reading, non-linguistic auditory SL (ASL), English visual SL (VSL), and Chinese ASL accounted for a significant amount of variance in English L2 word reading. Non-linguistic ASL, Chinese VSL, English VSL, and English ASL accounted for a significant amount of variance in Chinese word reading. Our results provide clear and novel evidence for cross-linguistic contribution from Chinese SL to English reading, and from English SL to Chinese reading, highlighting a bi-directional relationship between SL in one language and reading in another language.Constructing a student development model for undergraduate vocational universities in China using the Fuzzy Delphi Method and Analytic Hierarchy ProcessQiaona XingHuey Pyng TanSu Wan Gan10.1371/journal.pone.03010172024-03-22T14:00:00Z2024-03-22T14:00:00Z<p>by Qiaona Xing, Huey Pyng Tan, Su Wan Gan</p>
As the industrial structure changes, the severe shortage of high-quality technical and skilled talent in China is one of the most significant factors affecting the high-quality development of China’s economy. Bridging the gap between cultivating talent from new undergraduate vocational universities and the demand for industrial talent is regarded as an efficient strategy to address the talent shortage. In addressing the gap, China is hindered by a lack of clarity regarding student development goals and effective assessment instruments. Thus, this study aimed to use the Fuzzy Delphi Method (FDM) and the Analytical Hierarchy Process (AHP) to overcome the above challenges. Specifically, we used the FDM to establish a five-level undergraduate vocational education student development model with two 2<sup>nd</sup>-level indicators, three 3<sup>rd</sup>-level indicators, eight 4<sup>th</sup>-level indicators, and 33 5<sup>th</sup>-level indicators to clarify student development goals. Then, the AHP was applied to determine the indicator weights, and a student development assessment instrument was developed to help universities acquire student development data and improve the matching degree between talent supply and demand. This study could help undergraduate vocational universities cultivate high-quality technical and skilled talent quickly to meet the demand for China’s new economic system and to promote industry independence and global competitiveness.Application of a text mining method in navigation and communication for enhancing maritime safetyPaulina Hatłas-SowińskaLeszek Misztal10.1371/journal.pone.02995822024-03-22T14:00:00Z2024-03-22T14:00:00Z<p>by Paulina Hatłas-Sowińska, Leszek Misztal</p>
This paper introduces a model for the translation of natural language into ontology and vice versa in an autonomous navigation system of a sea-going vessel. The system comprehensively executes communication tasks at sea. The authors use machine learning methods in the field of text mining and basic and additional properties of ontologies. The newly developed ontology is applicable in shipping. The key elements of the prototype are the sequence of communication commands given from the ship’s bridge, decomposition, extraction of the communication sequence and the rule base. The presented model has been implemented and verified in selected scenarios of collision situations at sea. The test results confirm that the assumptions, the designed system architecture and the algorithms in the prototype are correct.Construction and improvement of English vocabulary learning model integrating spiking neural network and convolutional long short-term memory algorithmYunxia Wang10.1371/journal.pone.02994252024-03-22T14:00:00Z2024-03-22T14:00:00Z<p>by Yunxia Wang</p>
To help non-native English speakers quickly master English vocabulary, and improve reading, writing, listening and speaking skills, and communication skills, this study designs, constructs, and improves an English vocabulary learning model that integrates Spiking Neural Network (SNN) and Convolutional Long Short-Term Memory (Conv LSTM) algorithms. The fusion of SNN and Conv LSTM algorithm can fully utilize the advantages of SNN in processing temporal information and Conv LSTM in sequence data modeling, and implement a fusion model that performs well in English vocabulary learning. By adding information transfer and interaction modules, the feature learning and the timing information processing are optimized to improve the vocabulary learning ability of the model in different text contents. The training set used in this study is an open data set from the WordNet and Oxford English Corpus data corpora. The model is presented as a computer program and applied to an English learning application program, an online vocabulary learning platform, or a language education software. The experiment will use the open data set to generate a test set with text volume ranging from 100 to 4000. The performance indicators of the proposed fusion model are compared with those of five traditional models and applied to the latest vocabulary exercises. From the perspective of learners, 10 kinds of model accuracy, loss, polysemy processing accuracy, training time, syntactic structure capturing accuracy, vocabulary coverage, F1-score, context understanding accuracy, word sense disambiguation accuracy, and word order relation processing accuracy are considered. The experimental results reveal that the performance of the fusion model is better under different text sizes. In the range of 100–400 text volume, the accuracy is 0.75–0.77, the loss is less than 0.45, the F1-score is greater than 0.75, the training time is within 300s, and the other performance indicators are more than 65%; In the range of 500–1000 text volume, the accuracy is 0.81–0.83, the loss is not more than 0.40, the F1-score is not less than 0.78, the training time is within 400s, and the other performance indicators are above 70%; In the range of 1500–3000 text volume, the accuracy is 0.82–0.84, the loss is less than 0.28, the F1-score is not less than 0.78, the training time is within 600s, and the remaining performance indicators are higher than 70%. The fusion model can adapt to various types of questions in practical application. After the evaluation of professional teachers, the average scores of the choice, filling-in-the-blank, spelling, matching, exercises, and synonyms are 85.72, 89.45, 80.31, 92.15, 87.62, and 78.94, which are much higher than other traditional models. This shows that as text volume increases, the performance of the fusion model is gradually improved, indicating higher accuracy and lower loss. At the same time, in practical application, the fusion model proposed in this study has a good effect on English learning tasks and offers greater benefits for people unfamiliar with English vocabulary structure, grammar, and question types. This study aims to provide efficient and accurate natural language processing tools to help non-native English speakers understand and apply language more easily, and improve English vocabulary learning and comprehension.A fuzzy description logic based IoT framework: Formal verification and end user programmingMiguel Pérez-GasparJavier GomezEverardo BárcenasFrancisco Garcia10.1371/journal.pone.02966552024-03-22T14:00:00Z2024-03-22T14:00:00Z<p>by Miguel Pérez-Gaspar, Javier Gomez, Everardo Bárcenas, Francisco Garcia</p>
The Internet of Things (IoT) has become one of the most popular technologies in recent years. Advances in computing capabilities, hardware accessibility, and wireless connectivity make possible communication between people, processes, and devices for all kinds of applications and industries. However, the deployment of this technology is confined almost entirely to tech companies, leaving end users with only access to specific functionalities. This paper presents a framework that allows users with no technical knowledge to build their own IoT applications according to their needs. To this end, a framework consisting of two building blocks is presented. A friendly interface block lets users tell the system what to do using simple operating rules such as “if the temperature is cold, turn on the heater.” On the other hand, a fuzzy logic reasoner block built by experts translates the ambiguity of human language to specific actions to the actuators, such as “call the police.” The proposed system can also detect and inform the user if the inserted rules have inconsistencies in real time. Moreover, a formal model is introduced, based on fuzzy description logic, for the consistency of IoT systems. Finally, this paper presents various experiments using a fuzzy logic reasoner to show the viability of the proposed framework using a smart-home IoT security system as an example.Automatic detection of expressed emotion from Five-Minute Speech Samples: Challenges and opportunitiesBahman MirheidariAndré BittarNicholas CumminsJohnny DownsHelen L. FisherHeidi Christensen10.1371/journal.pone.03005182024-03-21T14:00:00Z2024-03-21T14:00:00Z<p>by Bahman Mirheidari, André Bittar, Nicholas Cummins, Johnny Downs, Helen L. Fisher, Heidi Christensen</p>
Research into clinical applications of speech-based emotion recognition (SER) technologies has been steadily increasing over the past few years. One such potential application is the automatic recognition of expressed emotion (EE) components within family environments. The identification of EE is highly important as they have been linked with a range of adverse life events. Manual coding of these events requires time-consuming specialist training, amplifying the need for automated approaches. Herein we describe an automated machine learning approach for determining the <i>degree of warmth</i>, a key component of EE, from acoustic and text natural language features. Our dataset of 52 recorded interviews is taken from recordings, collected over 20 years ago, from a nationally representative birth cohort of British twin children, and was manually coded for EE by two researchers (inter-rater reliability 0.84–0.90). We demonstrate that the degree of warmth can be predicted with an <i>F</i><sub>1</sub>-score of 64.7% despite working with audio recordings of highly variable quality. Our highly promising results suggest that machine learning may be able to assist in the coding of EE in the near future.A normalization model for repeated letters in social media hate speech text based on rules and spelling correctionZainab MansurNazlia OmarSabrina TiunEissa M. Alshari10.1371/journal.pone.02996522024-03-21T14:00:00Z2024-03-21T14:00:00Z<p>by Zainab Mansur, Nazlia Omar, Sabrina Tiun, Eissa M. Alshari</p>
As social media booms, abusive online practices such as hate speech have unfortunately increased as well. As letters are often repeated in words used to construct social media messages, these types of words should be eliminated or reduced in number to enhance the efficacy of hate speech detection. Although multiple models have attempted to normalize out-of-vocabulary (OOV) words with repeated letters, they often fail to determine whether the in-vocabulary (IV) replacement words are correct or incorrect. Therefore, this study developed an improved model for normalizing OOV words with repeated letters by replacing them with correct in-vocabulary (IV) replacement words. The improved normalization model is an unsupervised method that does not require the use of a special dictionary or annotated data. It combines rule-based patterns of words with repeated letters and the SymSpell spelling correction algorithm to remove repeated letters within the words by multiple rules regarding the position of repeated letters in a word, be it at the beginning, middle, or end of the word and the repetition pattern. Two hate speech datasets were then used to assess performance. The proposed normalization model was able to decrease the percentage of OOV words to 8%. Its F1 score was also 9% and 13% higher than the models proposed by two extant studies. Therefore, the proposed normalization model performed better than the benchmark studies in replacing OOV words with the correct IV replacement and improved the performance of the detection model. As such, suitable rule-based patterns can be combined with spelling correction to develop a text normalization model to correctly replace words with repeated letters, which would, in turn, improve hate speech detection in texts.Interpretable prediction of brain activity during conversations from multimodal behavioral signalsYoussef HmamoucheMagalie OchsLaurent PrévotThierry Chaminade10.1371/journal.pone.02843422024-03-21T14:00:00Z2024-03-21T14:00:00Z<p>by Youssef Hmamouche, Magalie Ochs, Laurent Prévot, Thierry Chaminade</p>
We present an analytical framework aimed at predicting the local brain activity in uncontrolled experimental conditions based on multimodal recordings of participants’ behavior, and its application to a corpus of participants having conversations with another human or a conversational humanoid robot. The framework consists in extracting high-level features from the raw behavioral recordings and applying a dynamic prediction of binarized fMRI-recorded local brain activity using these behavioral features. The objective is to identify behavioral features required for this prediction, and their relative weights, depending on the brain area under investigation and the experimental condition. In order to validate our framework, we use a corpus of uncontrolled conversations of participants with a human or a robotic agent, focusing on brain regions involved in speech processing, and more generally in social interactions. The framework not only predicts local brain activity significantly better than random, it also quantifies the weights of behavioral features required for this prediction, depending on the brain area under investigation and on the nature of the conversational partner. In the left Superior Temporal Sulcus, perceived speech is the most important behavioral feature for predicting brain activity, regardless of the agent, while several features, which differ between the human and robot interlocutors, contribute to the prediction in regions involved in social cognition, such as the TemporoParietal Junction. This framework therefore allows us to study how multiple behavioral signals from different modalities are integrated in individual brain regions during complex social interactions.Adaptation of the Client Diagnostic Questionnaire for East AfricaEdith Kamaru KwobahSuzanne GoodrichJayne Lewis KulzerMichael KanyesigyeSarah ObatsaJulius CheruiyotLorna KipronoColma KibetFelix OchiengElizabeth A. BukusiSusan OfnerSteven A. BrownConstantin T. YiannoutsosLukoye AtwoliKara Wools-Kaloustian10.1371/journal.pgph.00017562024-03-19T14:00:00Z2024-03-19T14:00:00Z<p>by Edith Kamaru Kwobah, Suzanne Goodrich, Jayne Lewis Kulzer, Michael Kanyesigye, Sarah Obatsa, Julius Cheruiyot, Lorna Kiprono, Colma Kibet, Felix Ochieng, Elizabeth A. Bukusi, Susan Ofner, Steven A. Brown, Constantin T. Yiannoutsos, Lukoye Atwoli, Kara Wools-Kaloustian</p>
Research increasingly involves cross-cultural work with non-English-speaking populations, necessitating translation and cultural validation of research tools. This paper describes the process of translating and criterion validation of the Client Diagnostic Questionnaire (CDQ) for use in a multisite study in Kenya and Uganda. The English CDQ was translated into Swahili, Dholuo (Kenya) and Runyankole/Rukiga (Uganda) by expert translators. The translated documents underwent face validation by a bilingual committee, who resolved unclear statements, agreed on final translations and reviewed back translations to English. A diagnostic interview by a mental health specialist was used for criterion validation, and Kappa statistics assessed the strength of agreement between non-specialist scores and mental health professionals’ diagnoses. Achieving semantic equivalence between translations was a challenge. Validation analysis was done with 30 participants at each site (median age 32.3 years (IQR = (26.5, 36.3)); 58 (64.4%) female). The sensitivity was 86.7%, specificity 64.4%, positive predictive value 70.9% and negative predictive value 82.9%. Diagnostic accuracy by the non-specialist was 75.6%. Agreement was substantial for major depressive episode and positive alcohol (past 6 months) and alcohol abuse (past 30 days). Agreement was moderate for other depressive disorders, panic disorder and psychosis screen; fair for generalized anxiety, drug abuse (past 6 months) and Post Traumatic Stress Disorder (PTSD); and poor for drug abuse (past 30 days). Variability of agreement between sites was seen for drug use (past 6 months) and PTSD. Our study successfully adapted the CDQ for use among people living with HIV in East Africa. We established that trained non-specialists can use the CDQ to screen for common mental health and substance use disorders with reasonable accuracy. Its use has the potential to increase case identification, improve linkage to mental healthcare, and improve outcomes. We recommend further studies to establish the psychometric properties of the translated tool.Call combination in African forest elephants <i>Loxodonta cyclotis</i>Daniela HedwigAnna Kohlberg10.1371/journal.pone.02996562024-03-18T14:00:00Z2024-03-18T14:00:00Z<p>by Daniela Hedwig, Anna Kohlberg</p>
Syntax, the combination of meaning-devoid phonemes into meaningful words, which in turn are combined in structurally and semantically complex sentences, is fundamental to the unlimited expressiveness of human languages. Studying the functions of call combinations in non-human species provides insights into the evolution of such syntactic capabilities. Here, we investigated the combination of high amplitude broadband calls with low frequency rumble vocalizations in a highly social species, the African forest elephant <i>Loxodonta cyclotis</i>. Rumbles play an integral role in coordinating social interactions by transmitting socially relevant information, including individual identity. By contrast, broadband calls, such as roars, are thought to function as signals of distress and urgency as they are typically produced in situations of high emotional intensity. Functional changes associated with the combination of these calls remain little understood. We found that call combinations were produced by all age-sex classes but were most prevalent in immature individuals. We found that rumbles used singularly occurred in all five investigated social contexts, whereas single broadband calls were restricted to two resource-related contexts. Call combinations also occurred in all five contexts, suggesting an increase in the functional use of broadband calls when combined with rumbles, analogous to the generativity brought about through syntax in human speech. Moreover, combining calls appeared to lead to functional shifts towards high-stake contexts. Call combinations were more likely in competition contexts compared to single rumbles, and more likely in separation contexts compared to single broadband calls. We suggest that call combination in forest elephants may aide to reduce message ambiguity in high-stake situation by simultaneously communicating distress and individual identity, which may be critical to secure access to resources, reduce the risk of injury and to reunite with or recruit the support of the family group.Can we spot fake public comments generated by ChatGPT(-3.5, -4)?: Japanese stylometric analysis expose emulation created by one-shot learningWataru ZaitsuMingzhe JinShunichi IshiharaSatoru TsugeMitsuyuki Inaba10.1371/journal.pone.02990312024-03-13T14:00:00Z2024-03-13T14:00:00Z<p>by Wataru Zaitsu, Mingzhe Jin, Shunichi Ishihara, Satoru Tsuge, Mitsuyuki Inaba</p>
Public comments are an important opinion for civic when the government establishes rules. However, recent AI can easily generate large quantities of disinformation, including fake public comments. We attempted to distinguish between human public comments and ChatGPT-generated public comments (including ChatGPT emulated that of humans) using Japanese stylometric analysis. Study 1 conducted multidimensional scaling (MDS) to compare 500 texts of five classes: Human public comments, GPT-3.5 and GPT-4 generated public comments only by presenting the titles of human public comments (i.e., zero-shot learning, GPT<sub>zero</sub>), GPT-3.5 and GPT-4 emulated by presenting sentences of human public comments and instructing to emulate that (i.e., one-shot learning, GPT<sub>one</sub>). The MDS results showed that the Japanese stylometric features of the public comments were completely different from those of the GPT<sub>zero</sub>-generated texts. Moreover, GPT<sub>one</sub>-generated public comments were closer to those of humans than those generated by GPT<sub>zero</sub>. In Study 2, the performance levels of the random forest (RF) classifier for distinguishing three classes (human, GPT<sub>zero</sub>, and GPT<sub>one</sub> texts). RF classifiers showed the best precision for the human public comments of approximately 90%, and the best precision for the fake public comments generated by GPT (GPT<sub>zero</sub> and GPT<sub>one</sub>) was 99.5% by focusing on integrated next writing style features: phrase patterns, parts-of-speech (POS) bigram and trigram, and function words. Therefore, the current study concluded that we could discriminate between GPT-generated fake public comments and those written by humans at the present time.Data-driven analytics for student reviews in China’s higher vocational education MOOCs: A quality improvement perspectiveHongbo LiHuilin GuXue HaoXin YanQingkang Zhu10.1371/journal.pone.02986752024-03-13T14:00:00Z2024-03-13T14:00:00Z<p>by Hongbo Li, Huilin Gu, Xue Hao, Xin Yan, Qingkang Zhu</p>
Higher vocational education is the core component of China’s national education system and shoulders the mission of cultivating high-skilled and applied talents. The wide application of Massive Open Online Courses (MOOCs) has effectively improved the curriculum system of China’s higher vocational education. In the meantime, some MOOCs suffer from poor course quality. Therefore, from the perspective of sustainable course quality improvement, we propose a data-driven framework for mining and analyzing student reviews in China’s higher vocational education MOOCs. In our framework, we first mine multi-level student demands hidden in MOOC reviews by combining web crawlers and text mining. Then we use an artificial neural network and the KANO model to classify the extracted student demands, thereby designing effective and sustainable MOOC quality improvement strategies. Based on the real data from China’s higher vocational education MOOCs, we validate the effectiveness of the proposed data-driven framework.Large language models are able to downplay their cognitive abilities to fit the persona they simulateJiří MiličkaAnna MarklováKlára VanSlambrouckEva PospíšilováJana ŠimsováSamuel HarvanOndřej Drobil10.1371/journal.pone.02985222024-03-13T14:00:00Z2024-03-13T14:00:00Z<p>by Jiří Milička, Anna Marklová, Klára VanSlambrouck, Eva Pospíšilová, Jana Šimsová, Samuel Harvan, Ondřej Drobil</p>
This study explores the capabilities of large language models to replicate the behavior of individuals with underdeveloped cognitive and language skills. Specifically, we investigate whether these models can simulate child-like language and cognitive development while solving false-belief tasks, namely, change-of-location and unexpected-content tasks. GPT-3.5-turbo and GPT-4 models by OpenAI were prompted to simulate children (N = 1296) aged one to six years. This simulation was instantiated through three types of prompts: plain zero-shot, chain-of-thoughts, and primed-by-corpus. We evaluated the correctness of responses to assess the models’ capacity to mimic the cognitive skills of the simulated children. Both models displayed a pattern of increasing correctness in their responses and rising language complexity. That is in correspondence with a gradual enhancement in linguistic and cognitive abilities during child development, which is described in the vast body of research literature on child development. GPT-4 generally exhibited a closer alignment with the developmental curve observed in ‘real’ children. However, it displayed hyper-accuracy under certain conditions, notably in the primed-by-corpus prompt type. Task type, prompt type, and the choice of language model influenced developmental patterns, while temperature and the gender of the simulated parent and child did not consistently impact results. We conducted analyses of linguistic complexity, examining utterance length and Kolmogorov complexity. These analyses revealed a gradual increase in linguistic complexity corresponding to the age of the simulated children, regardless of other variables. These findings show that the language models are capable of downplaying their abilities to achieve a faithful simulation of prompted personas.