Exploring linguistic fingerprints in human and AI-generated texts: An NLP-based approach in second language writing

Reading academic texts that include mainly university textbooks has been a challenge for EAP learners. There are various reasons for text difficulty; however, linguistic elements were investigated in this study. The aim of this study was to determine whether lexical richness of the readers would be a more potent predictor of their academic reading performance or their ability for producing and processing complex syntactic structures. The study involved 50 ELT teacher trainees, 25 juniors and 25 seniors, at Shahid Madani University of Azerbaijan, Iran. In a standard multiple regression design, the participants were given an opinion essay-writing task and an IELTS academic reading test. Their scores on IELTS academic reading test were regressed against LFP (Lexical Frequency Profile) and MLTU (Mean Length of T-Unit) indexes of their essays. LFP index is a measure of lexical richness adapted to the web for free online access under the name Web-VocabProfile, and MLTU index is a measure of syntactic complexity. Results indicated that the ability in producing and processing complex syntactic structures rather than mere grammatical knowledge can be considered as effective a predictor of academic reading comprehension as lexical richness. Therefore, lexical richness may no longer be supposed as the single most important predictor of academic reading performance.

This study investigated the patterns of noun phrase cohesion in English discourse, with a specific interest in the role of the different clause types. Using corpus methodology, the study synthesized into a single framework for analysis the central features of previous research regarding cohesive devices, preferred argument structure, genre, information packaging and clause structure. A corpus of 1206 noun phrases was coded for factors drawn from previous research, starting with whether or not the NP contained old/cohesive information. Results of frequency cross-tabulations and a factor analysis indicated that preferred argument structure, noun phrase form, and discourse genre were more significant influences on patterns of noun phrase cohesion in English discourse than clause structure. However, patterns of cohesive noun phrases according to the distance to their antecedents revealed that the more grammatically integrated clauses, such embedded infinitival clauses, the fewer cohesive noun phrases with antecedents in the immediate context they tended to have. This indicated that at the local inter-clausal level noun phrase cohesion and the level of grammatical integration of a combined clause existed in somewhat complementary distribution. Conclusions drawn included that clause grammar codes cohesion locally, displacing the need for noun phrases to mark cohesion in the immediate discourse. The study therefore quantitatively supports previous theories that discourse cohesion and the different types of combined clauses in English exist along a continuum from grammar to discourse.

The aim of this study was to examine the possible link between linguistic accuracy and syntactic complexity in written skills in a sample of native Greek speakers who are learning Spanish as a foreign language. Considering the syntactic complexity as an index of interest in addressing written production a comparison between the native and nonnative written production was also pursued with regards to the degree of syntactic elaboration. To this end, a narrative task under time pressure was used to measure writing skills while the evaluation of the performance was based on the quantification of two linguistic dimensions: linguistic accuracy and syntactic complexity. The statistical analyses showed that there was no significant association between linguistic accuracy and syntactic complexity which seems to refute the existence of trade off. Moreover, it was observed that native speakers seem to use a more elaborate syntax in their texts, a result that was only partially true if one also takes into account the implications of other factors in the results obtained. The results are discussed based on the postulates of the Cognition Hypothesis and the Tradeoff Hypothesis.

Academic writing has played an essential role in communicating the cognitive aspects of the human mind. Natural Language Processing (NLP) tools enable us to examine linguistic knowledge. However, writing patterns and applicable linguistic characteristics differ geographically. The study's primary purpose is to understand the global writing pattern and linguistic diversities of research articles in the LIS domain. The corpus was identified from four SCOPUS-enrolled open-access libraries and information science journals. The journals published in India and outside India were selected for the study in 2020. The syntactic complexity in 147 text documents was measured using the Tool for the Automatic Analysis of Syntactic Sophistication and Complexity (TASSAC). The corpus was further examined using the Structural Equation Model (SEM) to determine the causal relationship among independent variables such as syntax features and readability scores. The results depict the differences in the patterning of syntactic features at both the global and national levels. Furthermore, the study allows us to see how linguistic diversity is underplayed in research writings and helps to understand writing patterns through cross-country comparisons. Furthermore, the paper employs model-based reasoning to identify global and national latent variables.

This study aimed to examine how speech while sign (simultaneous communication [SimCom]) affects the spoken language of bimodal bilingual teachers and how individual differences in sign-language vocabulary knowledge, SimCom teaching experience, and the ability to perform speech under dual-task conditions explain the variability in SimCom performance. Forty experienced teachers of deaf and hard of hearing students participated in a story narration task under different conditions. Speech rate, lexical richness, and syntactic complexity were measured and compared across speech-only versus SimCom conditions. Furthermore, participants' score on a sign-language vocabulary test, their self-reported SimCom teaching experience, and their performance in a dual-task condition were taken as predictors of SimCom narration performance. The findings revealed slower speech rate, lower lexical richness, and lower syntactic complexity in the SimCom condition compared with the speech-only condition. Sign-language vocabulary score and SimCom teaching experience explained speech rate and lexical richness. Participant's ability to speak under a dual-task condition did not modulate performance. The findings may suggest that the production of the less dominant (sign) language during SimCom entails inhibition of the dominant (spoken) language relative to the speech-only condition. At the same time, the findings are also compatible with the suggestion that SimCom serves as a unique complex communication unit that cannot be reduced to the combination of two languages.

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

This study employs the Python-based LFTK(Linguistic Feature Toolkit) to explore the linguistic characteristics of compositions by English learners from ten Asian countries, collected in the ICNALE(International Corpus Network of Asian Learners of English). The present study investigates linguistic peculiarities including sentence structures, syntactic complexity, lexical readability, and lexical richness while focusing on the differences between original compositions and their edited versions across various proficiency levels. The results reveal that the linguistic features are closely connected with the learners' proficiency levels; the edited versions are to show improved syntactic complexity, lexical readability, and lexical richness of the original compositions Finally, this paper may shed light on effective teaching methodology to English learners in Asia across various educational contexts.

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

Poetry summarization presents unique challenges for Large Language Models due to its dense use of metaphor, imagery, rhythm, and stylistic elements. This study analyzes the zero-shot techniques for poem summarization using the five state-of-the-art Large Language Models (LLMs), such as Gemini 2.5 Flash, Groq Mini, DistilBART CNN-12-6, LLaMA 3.3 70B Versatile, and Moonshot AI Kimi K2 Instruct. This study also evaluated the summaries using ROUGE-1, ROUGE-2, ROUGEL, and BERTScore to measure both lexical overlap and semantic similarity. After that, use a novel LLM-as-a-Judge approach where evaluate summaries on content similarity, quality, completeness, accuracy, and overall match, and average score for each of the LLMs. Experimental results show that DistilBART CNN12-6 achieves the highest ROUGE and BERTScore performance, while Gemini 2.5 Flash achieves the best performance in LLM-as-a-judge evaluation. This study provides a clear and scalable way to evaluate poem summarization. Overall, this study shows that zero-shot techniques in LLMs can effectively generate high-quality poem summaries.

P03-104 - Linguistic production and comprehension deficits in schizophrenia and bipolar disorder

A textual data processing task that involves the automatic extraction of relevant and salient keyphrases from a document that expresses all the important concepts of the document is called keyphrase extraction. Due to technological advancements, the amount of textual information on the Internet is rapidly increasing as a lot of textual information is processed online in various domains such as offices, news portals, or for research purposes. Given the exponential increase of news articles on the Internet, manually searching for similar news articles by reading the entire news content that matches the user’s interests has become a time-consuming and tedious task. Therefore, automatically finding similar news articles can be a significant task in text processing. In this context, keyphrase extraction algorithms can extract information from news articles. However, selecting the most appropriate algorithm is also a problem. Therefore, this study analyzes various supervised and unsupervised keyphrase extraction algorithms, namely KEA, KP-Miner, YAKE, MultipartiteRank, TopicRank, and TeKET, which are used to extract keyphrases from news articles. The extracted keyphrases are used to compute lexical and semantic similarity to find similar news articles. The lexical similarity is calculated using the Cosine and Jaccard similarity techniques. In addition, semantic similarity is calculated using a word embedding technique called Word2Vec in combination with the Cosine similarity measure. The experimental results show that the KP-Miner keyphrase extraction algorithm, together with the Cosine similarity calculation using Word2Vec (Cosine-Word2Vec), outperforms the other combinations of keyphrase extraction algorithms and similarity calculation techniques to find similar news articles. The similar articles identified using KPMiner and the Cosine similarity measure with Word2Vec appear to be relevant to a particular news article and thus show satisfactory performance with a Normalized Discounted Cumulative Gain (NDCG) value of 0.97. This study proposes a method for finding similar news articles that can be used in conjunction with other methods already in use.

It is debated whether cortical responses matching the time scales of phrases and sentences mediate the mental construction of the syntactic chunks or are simply caused by the semantic properties of words. Here, we investigate to what extent delta-band neural responses to speech can be explained by semantic relatedness between words. To dissociate the contribution of semantic relatedness from sentential structures, participants listened to sentence sequences and paired-word sequences in which semantically related words repeated at 1 Hz. Semantic relatedness in the 2 types of sequences was quantified using a word2vec model that captured the semantic relation between words without considering sentential structure. The word2vec model predicted comparable 1-Hz responses with paired-word sequences and sentence sequences. However, empirical neural activity, recorded using magnetoencephalography, showed a weaker 1-Hz response to paired-word sequences than sentence sequences in a word-level task that did not require sentential processing. Furthermore, when listeners applied a task-related rule to parse paired-word sequences into multi-word chunks, 1-Hz response was stronger than that in word-level task on the same sequences. Our results suggest that cortical activity tracks multi-word chunks constructed by either syntactic rules or task-related rules, whereas the semantic relatedness between words contributes only in a minor way.

Abstract: The purpose of this study is to analyze the progress of 14 Spanish-speaking learners of English during a period abroad from a longitudinal perspective. Oral and written data were collected three times during an academic year at a British university. These samples were analyzed in terms of fluency, syntactic complexity, lexical richness, and accuracy. The results of the statistical analyses indicate that, while a few months abroad might be sufficient for some gains in oral performance to occur, improvement in written production is slower and does not seem to take place until students have spent more than one semester abroad. In addition, it was observed that the type of interaction experienced abroad and some attitudinal features can partly explain language development in some areas.

This study examined whether AI-based editing software can contribute to the improvement of L2 learners’ writing quality measured by four indicators: syntactic complexity, lexical variety, lexical richness, and readability. A total of 65 college students at a low-intermediate level of fluency participated in this study. One class, consisting of 30 students, was randomly designated as the control group, and another class with 35 students as the experimental group. Free versions of Grammarly, an AI-based written corrective feedback provider, and Quillbot, an AI-based paraphrasing suggestion provider, were used as a treatment for the experimental group. The analysis revealed that the experimental group outperformed the control group in terms of syntactic complexity, lexical variety, and readability. However, little difference has been detected between the groups in terms of lexical richness. This result suggests potential pedagogical capabilities AI-based editing software can offer L2 writers and teachers.

This study adopts a mixed‐methods design to systematically explore the impact of L1 and L2 use in prewriting discussions on L2 writing quality, cognitive feedback, and social interaction of intermediate Chinese as a foreign language (CFL) learners. Eighty intermediate‐level CFL learners from Southeast Asia participated in two experiments, which comprehensively evaluated their written output across six dimensions: lexical complexity, syntactic complexity, accuracy, fluency, content richness, and holistic quality. The results indicated that discussions in L1 significantly improved content richness, while discussions in L2 significantly enhanced lexical complexity, fluency, and holistic quality. However, no significant effects were found in syntactic complexity and accuracy between the two language conditions. Participants' feedback further revealed that L1 discussions facilitated content generation, while L2 discussions notably promoted fluency and holistic quality improvement. Despite some discrepancies between actual writing performance and participants' perceived feedback, the overall findings suggested that L2 discussions had a more significant impact on improving writing quality for intermediate CFL learners.