Word Substitution Research Articles

Effects of Automated Feedback on English as a Foreign Language Learners’ Writing Performance: Evidence from a Quasi-experiment

Automated writing evaluation (AWE) is increasingly used to provide formative feedback on second language (L2) students’ writing. A key factor influencing the effectiveness of AWE feedback on L2 writing performance is the learners’ revision behaviors as they process the feedback. Adopting a quasi-experiment, this study aims to evaluate the impacts of Criterion automated corrective feedback (ACF) on English as a foreign language (EFL) students’ writing performance based on two measures of accuracy: overall writing accuracy and accuracy of English article usage. Learners’ textual operations in response to Criterion ACF were examined for possible explanations for recorded gains (if any) in their writing accuracy. The main findings indicate a lack of intervention and retention effects on learners’ accuracy over the semester during which Criterion ACF was incorporated to supplement the writing instructor's feedback on organization and content. In addition, across four writing entries conducted on Criterion, learners’ revisions to their essays following Criterion ACF were primarily at the local level, dominated by addition, deletion, or substitution of individual words or short phrases rather than substantive revisions to their scripts. About one third of all Criterion feedback points did not result in textual changes to the first drafts, indicating a moderate uptake rate of the feedback. Implications related to formative feedback practices in the EFL writing classroom and the adaptation of Criterion's technical capacities are accordingly presented.

Unveiling tone manipulation in MD&A: Evidence from ChatGPT experiments

This research uses ChatGPT-3.5 and ChatGPT-4 to investigate tone manipulation in the management discussion and analysis of annual reports of Chinese-listed companies. We find that the quantification of emotional content in text using financial BERT model and dictionary approach is inconsistent due to two kinds of manipulation: "Expression Manipulation" and "Word Manipulation". Based on ChatGPT we verify the manipulative behavior of complex expressions and sentiment word substitutions. Our research suggests that using ChatGPT with appropriate cue words can help alleviate tone manipulation in financial texts, with GPT-4 having a better effect.

Enhancing Contextual Masking in Reversible Linguistic Steganography with Ensemble Methods

Various cybercrimes can be prevented by text authentication that is responsible for preserving digital identities and contents. Digital signatures come in handy as a way of authenticating texts, which is an extensively used method. One approach to this problem is linguistic steganography, which allows hiding the signature in other words within the text and thereby facilitating efficient data management. However, it should be noted that there is a danger that these kinds of changes may result in inappropriate decisions being taken by automated computing systems not to mention change their final outputs (unseen). As such, many people are becoming more concerned with the possibility of reversing steganography so that it becomes possible to eliminate any distortions made during the process. This paper uses Contextual masking instead of masking randomly with BERT model. The goal behind this research was developing a natural language text specific Reversible Steganographic System. Our model uses pre-trained BERT as a transformer based masked language model and reversibly embeds messages through predictive word substitution. To quantify predictive uncertainty, we introduce an adaptive steganographic technique using Bayesian deep learning. This experiment shows us how our proposed system balances imperceptibility with capacity while maintaining near semantics at all times. Also, we integrate ensemble methods instead of Monte Carlo to balance the imperceptibility.

Weight Saliency search with Semantic Constraint for Neural Machine Translation attacks

Text adversarial attack is an effective way to improve the robustness of Neural Machine Translation (NMT) models. Existing NMT attack tasks are often completed by replacing words. However, most of previous works pursue a high attack success rate but produce semantic inconsistency sentences, leading to wrong translations even for humans. In this paper, we propose a Weight Saliency search with Semantic Constraint (WSSC) algorithm to make semantic consistency word modifications to the input sentence for black-box NMT attacks. Specifically, our WSSC has two major merits. First, it optimizes the word substitution with a word saliency method, which is helpful to reduce word replacement rate. Second, it constrains the objective function with a semantic similarity loss, ensuring every modification does not lead to significant semantic changes. We evaluate the effectiveness of the proposed WSSC by attacking three popular NMT models, i.e., T5, Marian, and BART, on three widely used datasets, i.e., WMT14, WMT16, and TED. Experimental results validate that our WSSC improves Attack Success Rate (ASR) by 4.02% and Semantic Similarity score (USE) by 1.28% on average. Besides, our WSSC also shows good properties in keeping grammar correctness and transfer attack.

Stameering Speech Signal Segmentation and Classification using Machine Learning

Stuttering or Stammering is a speech defect within which sounds, syllables, or words are rehashed or delayed, disrupting the traditional flow of speech. Stuttering can make it hard to speak with other individuals, which regularly have an effect on an individual's quality of life. Automatic Speech Recognition (ASR) system is a technology that converts audio speech signal into corresponding text. Presently ASR systems play a major role in controlling or providing inputs to the various applications. Such an ASR system and Machine Translation Application suffers a lot due to stuttering (speech dysfluency). Dysfluencies will affect the phrase consciousness accuracy of an ASR, with the aid of increasing word addition, substitution and dismissal rates. In this work we focused on detecting and removing the prolongation, silent pauses and repetition to generate proper text sequence for the given stuttered speech signal. The stuttered speech recognition consists of two stages namely classification using ANN and testing in ASR. The major phases of classification system are Re-sampling, Segmentation, Pre Emphasis, Epoch Extraction and Classification. The current work is carried out in UCLASS Stuttering dataset using MATLAB with 4% to 6% increase in accuracy by ANN.

President of the Republic of Belarus – head or head of state?

Using linguistic and historical and legal methods, the author set out to show that the President of the Republic of Belarus is not the head of state, but the head of state. It is noted that the constitutional content of the concept of “head of state” does not significantly depend on the form of government. The issues of incorrect translation and substitution of words in аrt. 79 of the draft Constitution of the Republic of Belarus at the republican referendum on November 24, 1996. As a result, the President of the Republic of Belarus unreasonably, in addition to the will of the people, received the status of ,,Head of State”. An assessment of the consequences for the country and society of the ,,President-Head of State” union is given. It is shown that such a combination allows the President not to act, but to rule alone, which leads to the formula: ,,president” – ,,head of state” – ,,sovereign”. In conclusion, it is concluded about the monarchization of the institution of the presidency and the formation of a new atypical form of government in the country – the ,,republican elected monarchy”.

Biomedical text readability after hypernym substitution with fine-tuned large language models.

The advent of patient access to complex medical information online has highlighted the need for simplification of biomedical text to improve patient understanding and engagement in taking ownership of their health. However, comprehension of biomedical text remains a difficult task due to the need for domain-specific expertise. We aimed to study the simplification of biomedical text via large language models (LLMs) commonly used for general natural language processing tasks involve text comprehension, summarization, generation, and prediction of new text from prompts. Specifically, we finetuned three variants of large language models to perform substitutions of complex words and word phrases in biomedical text with a related hypernym. The output of the text substitution process using LLMs was evaluated by comparing the pre- and post-substitution texts using four readability metrics and two measures of sentence complexity. A sample of 1,000 biomedical definitions in the National Library of Medicine's Unified Medical Language System (UMLS) was processed with three LLM approaches, and each showed an improvement in readability and sentence complexity after hypernym substitution. Readability scores were translated from a pre-processed collegiate reading level to a post-processed US high-school level. Comparison between the three LLMs showed that the GPT-J-6b approach had the best improvement in measures of sentence complexity. This study demonstrates the merit of hypernym substitution to improve readability of complex biomedical text for the public and highlights the use case for fine-tuning open-access large language models for biomedical natural language processing.

A reversible natural language watermarking for sensitive information protection

Existing methods have evolved from using synonym substitution to incorporating arbitrary word substitution to achieve reversible natural language watermarking. However, a notable limitation is that they are prone to overlook the sensitivity of information associated with the original words, with a tendency to prefer non-sensitive words for substitution. As a result, a potential risk of sensitive information leakage contained in the original text is posed. Furthermore, while aiming for reversibility, the overall performance of the watermarking method may be inadvertently compromised. In response to the above problems, this paper puts forward a novel reversible natural language watermarking method that combines a Keyword Substitution scheme and a Prediction Error Expansion algorithm (KSPEE) to protect sensitive information, verify content integrity, protect copyright, and so on. Specifically, KSPEE leverages a keyword extraction algorithm to identify important content containing sensitive information in the original text, thereby determining the potential positions for watermark information embedding. Subsequently, a masked language model is utilized to predict appropriate substitution words based on the surrounding semantic information of the embedding position. In addition, the prediction error expansion algorithm is employed to select appropriate words for substituting the original keywords, ensuring the successful embedding of watermark information while maintaining the recoverability of the original keywords. By identifying keywords and substituting them, a suitable method of protecting the original sensitive information is provided. Extensive experiments demonstrate that, under the promise of semantic distortion and lossless restoration of the original content, the proposed method KSPEE achieves outstanding watermarked text quality. A higher watermark embedding rate is achieved and strong security is shown by KSPEE. More importantly, KSPEE effectively prevents the leakage of sensitive information.

Emojis in EFL Communication: An Analysis оf the Language-Like Functions in Emoji Use by Saudi Female Learners

Emojis are fast becoming an integral part of communication and taking the shape of a ‘language’ in its own right. This raises pertinent questions as to the future look of formal digital communication and the purposes for which this visual input is used by EFL learners. This mixed methods study investigated what prompts female EFL learners at Qassim University to use emojis, the communicative functions they fulfill, and whether their use hinders or facilitates communication. A WhatsApp group was created. A total of 21 female students were added to the group. The researcher observed their chat for 4 weeks. An interview was also used to collect data on the purposes and functions of using emojis. Results indicate that there are nine broad functions that are fulfilled by emojis when used by EFL learners: expression of emotions, enhancing tone, visual quotient, regional variations, efficient communication, creativity, word substitution, peer prompt, and playfulness. Further, irrespective of the purpose of emoji use, the participants see it as an enhancement of their English communication on a formal platform, and that fears that visuals can fully replace the text for foreign language learners are totally unfounded. This study recommends the investigation of using emojis for developing students' autonomy.

Speakers Raise Their Hands and Head During Self-Repairs in Dyadic Conversations

People often encounter difficulties in building shared understanding during everyday conversation. The most common symptom of these difficulties are self-repairs, when a speaker restarts, edits or amends their utterances mid-turn. Previous work has focused on the verbal signals of self-repair, i.e. speech disfluences (filled pauses, truncated words and phrases, word substitutions or reformulations), and computational tools now exist that can automatically detect these verbal phenomena. However, face-to-face conversation also exploits rich non-verbal resources and previous research suggests that self-repairs are associated with distinct hand movement patterns. This paper extends those results by exploring head and hand movements of both speakers and listeners using two motion parameters: height (vertical position) and 3D velocity. The results show that speech sequences containing self-repairs are distinguishable from fluent ones: speakers raise their hands and head more (and move more rapidly) during self-repairs. We obtain these results by analysing data from a corpus of 13 unscripted dialogues, and we discuss how these findings could support the creation of improved cognitive artificial systems for natural human-machine and human-robot interaction.

Resilience among young mothers affected by HIV in South Africa: Adaptations and psychometric properties of the Child and Youth Resilience Measure-Revised (CYRM-R) in a large cohort

BackgroundWhile growing evidence has documented risks and resilience to HIV and poor sexual and reproductive health (SRH) among young women, young mothers’ abilities to respond to and overcome challenges have been understudied. Developing appropriate methodologies to measure resilience in the context of HIV, especially among young mothers, is a critical first step to filling this gap. We aimed to improve the relevance, and comprehensibility of an open-access resilience measure for use among young mothers living in HIV-endemic communities in South Africa. This paper summarizes a two-stage study that was carried out in the Eastern Cape, South Africa. MethodsFirst, n = 9 cognitive interviews were conducted to inform the adaptation of the Child and Youth Resilience Measure Revised (CYRM-R) for use among young mothers (n = 7 living with HIV). The adapted measure was applied in remote interviews with 892 adolescent mothers during the COVID-19 pandemic (December 2021–April 2023). We investigated the basic psychometric properties of the adapted measure using STATA16. ResultsCognitive interviews offered unique perspectives on how to modify items in English and isiXhosa, including items that had similar understandings and responses, with new probes and additional key words added to seven items. Participants' understanding of item wording and the translation of items resulted in word changes or substitution to maximise participants’ understanding of eight out of 17 items. The utilization of vignettes proved to be useful in engaging participants to reflect on their experiences. The revised scale had moderate psychometric properties. The EFA confirmed a two-factor structure, and the internal consistency of the CYRM-R was moderate (α = 0.75; ω = 0.75). ConclusionCognitive interviews supported the adaptation of a resilience measure for adolescent mothers affected by HIV in South Africa. Initial psychometric properties highlight possible further adaptations needed, potentially due to the unique population of adolescent mothers.

Semantic-Preserving Adversarial Text Attacks

Deep learning models are known immensely brittle to adversarial text examples. Existing text adversarial attack strategies can be roughly divided into character-level, word-level, and sentence-level attacks. Despite the success brought by recent text attack methods, how to induce misclassification with minimal text modifications while keeping the lexical correctness, syntactic soundness, and semantic consistency is still a challenge. In this paper, we devise a Bigram and Unigram-based adaptive Semantic Preservation Optimization (BU-SPO) approach which attacks text documents not only at a unigram word level but also at a bigram level to avoid generating meaningless sentences. We also present a hybrid attack strategy that collects substitution words from both synonyms and sememe candidates, to enrich the potential candidate set. Besides, a Semantic Preservation Optimization (SPO) method is devised to determine the word substitution priority and reduce the perturbation cost. Furthermore, we constrain the SPO with a semantic Filter (dubbed SPOF) to improve the semantic similarity. To estimate the effectiveness of our proposed methods, BU-SPO and BU-SPOF, we attack four victim deep learning models trained on three text datasets. Experimental results demonstrate that our approaches accomplish the highest semantics consistency and attack success rates by making minimal word modifications compared with competitive methods.

Lexical features of Russian speech of bilinguals in Germany and monolinguals in Russia: an experimental study

The results of the comparison of lexical features of Russian speech of four groups of respondents are presented: 1) adult Russian-German bilinguals aged 35-50 who moved to Germany in the 1990-2010s; 2) their children aged 10-15 who were born in Germany or moved to Germany at an early age; 3) adult monolinguals aged 35-50 living in St. Petersburg; 4) their children aged 10-15. The relevance of the research is, on the one hand, in the importance of studying the state of the Russian language in the families of Russian compatriots living abroad, its preserving and developing, and on the other hand, in the need to supplement the existing data on the speech development of bilinguals with new facts. The research is aimed at comparing lexical features of Russian speech of two generations of bilinguals in Germany and monolinguals in Russia. The material of the research includes transcripts of picture story recordings from the book of M. Mayer “Frog, where are you?”. The methods of the research are observation, data systematization and statistical processing, comparison, quantitative and qualitative interpretation of data. The authors found out the average proportion of lexical norms violations in the stories of informants and among them the proportion of word substitutions, word omissions and superfluous words insertion. The types of word substitutions, their percentage, and their reasons were determined. The similarity of lexical norms violations in the speech of children (bilingual and monolingual), conditioned by general laws of speech development, was revealed. The conclusion is made about the relatively stable Russia lexical system in the diaspora, at least in the two groups of Russian-German bilinguals studied, and about its similarity with the lexical system of monolinguals. Some parts of the lexical system of the Russian language of bilingual children aged 10-15 years undergo changes, but these changes do not violate its integrity.

Transfer Learning-Based Neural Machine Translation for Low-Resource Languages

College of Liberal Arts, Ludong University, Yantai 264025, China Chinese Lexicography Research Center, Yantai 264025, China Neural Machine Translation (NMT) improves readability by augmenting sentence suggestions based on the precise likelihood of the words. The word suggestions are trained using learning paradigms through repeated translations, word searches, and user inputs. However, the challenging process is the implication of NMT for low-resource language wherein the chances of false suggestions/ word substitutions are high. So, this article proposes a Likelihood-based Machine Translation Model (LMTM) for low-resource languages. The model uses word frequency and potential substitutions from less-known sentences to identify sentences with high precision. This is achieved through a combination of recurrence and substitutions using transfer learning. The identified high-likelihood words are used for sentence augmentation, and the entire set of words from the generated sentence updates the learning paradigm. The model suggests the highest likelihood words for NMT, preventing sentence falsification and ensuring accurate translations. The proposed model increases likelihood by 9.15%, correctness by 7.46%, and substitutions by 7.7%, respectively. It reduces falsification and time complexity by 9.33% and 8.52%, respectively. Overall, the LMTM improves translation quality for low-resource languages.

Performance analysis of robustness of BERT model under attack

With the aim of testing the robustness of machine learning models, this paper tests the performance of five classification models based on IMDB datasets. Furthermore, in this work, two types of sentence embeddings generated by word2vec and BERT are added with the noise of the normal distribution with different intensities. They are fed into the Support Vector Machine for testing. The experimental results show that the performance of the model slowly decreases as the noise intensity is increased. BERT-based sentiment embedding reduces less than Word2vec-based sentiment embedding. This paper believes that this experimental phenomenon can be used as a basis to support the robustness of BERT representation learning. At the same time, this paper also noted that the previous work experimented with perturbating the BERT model by replacing the words in the original text of the IMDB dataset, causing the BERT model performance to drop sharply. However, the experiment in this paper tested the robustness of the BERT model by embedding tampering, and the experimental results were stable. In this paper, it is assumed that word substitution and embedding interference are possible situations when operating the model. Therefore, for the two different experimental phenomena, this paper plans to design more detailed experiments in the next work to check the robustness of the model.

On English proverb variation from the perspective of linguistic creativity.

Proverbs are usually regarded as structurally fixed expressions. However, in daily communication, language users often change them to suit their communicative purposes in many ways, resulting in proverb variations. Using the data from the Corpus of Contemporary American English (COCA corpus), this study attempts to present varieties of the English proverb "There are two sides to every coin" and explain the variations from the perspective of linguistic creativity. This study also explores the variations of this proverb in EFL learners' use via the data from Chinese EFL learners' corpus TECCL. The study shows that, first proverb use can roughly be divided into two types: canonical and non-canonical uses, each having three ways of alteration, i.e., addition of modifiers, substitution of content words, and reduction. Second, Chinese EFL learners tend to use the proverb in a mechanical way with little variation, which shows their inflexible use of proverbs. Finally, proverb variation by nature is the creative manipulation of language use to fit the context, which is a form of linguistic creativity that reflects the cognitive creativity of human beings.

اِختلافِ مراجعِ ضمائراورقرآنی اُصولِ یُسر (مفسرین کی آراء کا ناقدانہ جائزہ)

It is impossible to receive guidance from the Holy Quran unless it is properly understood by its reader with the feeling that the Quran is descending upon his heart and soul. Notably, the Holy Quran cannot be understood without its interpretation and exposition. It is a common practice among linguists to use pronouns to create collaboration and correspondence among words and sentences. Pronouns are the substitutions of the original words and are used to avoid repetitions and make the statement more beautiful and eye-catching. The style of expression of the holy Quran is so simple that the pronouns, similes, examples, metaphors, and reasoning present in it, offered no difficulty even to its initial addressees who were quite illiterate people. The Holy Quran is the eternal code of conduct for the people. Everyone can seek guidance for every walk of life from it and this is the miracle of the holy Quran the any Muslim can memorize it with a little effort, whether he is an Arab or non-Arab. This is so because there is no ambiguity or complexity in it. This paper presents an analytical study of the pronouns to reinforce the Quranic princples of the easement.

Supremacy of suprasegmentals in Arabic phonology: Evidence from malapropisms

Speech errors are an important source of information to understand language processing and production. Earlier research focused on different types of errors including semantic and phonological errors while malapropisms, which refer to slips of the tongue involving whole word substitutions that share phonological similarities but are not related semantically, have not received adequate attention in the Arabic language. Drawing on malapropisms in Jordanian Arabic, we bring evidence on the supremacy of suprasegmental phonological aspects in Arabic phonology. This is unexpected as stress in Arabic is non-phonemic and fully predictable, besides Arabic rhythm is much less stress-timed than that of Germanic languages. Data was collected from spontaneous speech over a period of three years. Results showed that malapropisms share the primary stress position, the number of syllables and the word rhythmic pattern with the target words. To a lesser degree, the target and the error share the same rime and initial segments. Findings suggest that suprasegmental features are very crucial in Arabic phonology, like in Indo-European languages. Evidence suggests that formal similarity that is based on the syllabic and metrical structure of words plays a significant role in language processing and the organization of the mental lexicon in Arabic, which suggests that this is a language universal. Furthermore, our findings do not agree with earlier claims that Arabic has a flat syllabic structure. Rather, evidence suggests that Arabic, like English, has a hierarchical syllable structure, which seems to represent another language universal. More research on other Arabic dialects is recommended to corroborate these findings. Disclosure statement No potential conflict of interest was reported by the authors. * Corresponding author: Mohammed Nour Abu Guba, 0000-0002-5007-6439 mabu-gub@sharjah.ac.ae

Summarization Attack via Paraphrasing (Student Abstract)

Many natural language processing models are perceived to be fragile on adversarial attacks. Recent work on adversarial attack has demonstrated a high success rate on sentiment analysis as well as classification models. However, attacks to summarization models have not been well studied. Summarization tasks are rarely influenced by word substitution, since advanced abstractive summary models utilize sentence level information. In this paper, we propose a paraphrasing-based attack method to attack summarization models. We first rank the sentences in the document according to their impacts to summarization. Then, we apply paraphrasing procedure to generate adversarial samples. Finally, we test our algorithm on benchmarks datasets against others methods. Our approach achieved the highest success rate and the lowest sentence substitution rate. In addition, the adversarial samples have high semantic similarity with the original sentences.

Generation-based parallel particle swarm optimization for adversarial text attacks

Text adversarial attack is an effective strategy to investigate the vulnerability of Natural Language Processing (NLP) models. Most of text attack studies focus on word-level attacks with static or dynamic optimization algorithms. However, they are hard to balance (1) attack performance (i.e., attack success rate, word substitution rate) and (2) attack efficiency. Generally, static optimization is fast but suffers from low attack performance, and the dynamic adversary improves the attacking quality but is time-consuming. To address these challenges, a Generation-based Parallel Particle Swarm Optimization (GP2SO) is proposed for the adversarial text attack. Specifically, the GP2SO employs an adaptive strategy to determine the word modification priority, which produces a high attack performance owing to the aggressive objective function. To achieve time efficiency, we parallelize the PSO on multiple pipelines in a generation-overlapping manner. Extensive experiments on four public text recognition datasets are conducted by attacking four deep models to evaluate the effectiveness of the GP2SO. Experimental results manifest that the proposed GP2SO averagely improves the time efficiency by 272% with only 0.3% success rate reduction compared to the PSO. Besides, the GP2SO also shows superiorities in adversarial training and transferability compared with baselines. The code is provided to ensure reproducibility https://github.com/OutdoorManofML/GPPSO.