Electronic Health Records (EHRs) have revolutionised healthcare by enhancing patient care and facilitating provider communication. Nevertheless, the efficient extraction of valuable information from EHRs poses challenges, primarily due to the overwhelming volume of unstructured data, the wide variability in data formats, and the lack of standardised labels. Leveraging deep learning and concept embeddings, we address the gap in context-aware systems for EHRs. The proposed solution was evaluated on the MIMIC III dataset and demonstrated superior performance compared to other methodologies. We addressed the positive impact of the latent feature combined with the note representation in four different settings. Model performance was evaluated using a case study conducted with BertScore, assessing precision, recall, and F1 scores. The model excels in Medical Natural Language Inference (MedNLI) with an 89.3% accuracy, further boosted to 90.5% through retraining the embeddings using International Classification of Diseases (ICD) codes, which we formally designate as ClinicNarrIR. The ClinicNarrIR was tested with 1000 randomly sampled notes, achieving an N DCG@10 score of approximately 0.54 with accuracy@10 of 0.85. The study also demonstrates a high correlation between the results produced by the proposed representation and medical coders. Notably, in all evaluation cases, the optimal base pretrained model that emerged was BlueBERT.

It is neither possible nor fair to compare the performance of question-answering systems for the Holy Quran and Hadith Sharif in Arabic due to both the absence of a golden test dataset on the Hadith Sharif and the small size and easy questions of the newly created golden test dataset on the Holy Quran. This article presents two question–answer datasets: Hadith Question–Answer pairs (HAQA) and Quran Question–Answer pairs (QUQA). HAQA is the first Arabic Hadith question–answer dataset available to the research community, while the QUQA dataset is regarded as the more challenging and the most extensive collection of Arabic question–answer pairs on the Quran. HAQA was designed and its data collected from several expert sources, while QUQA went through several steps in the construction phase; that is, it was designed and then integrated with existing datasets in different formats, after which the datasets were enlarged with the addition of new data from books by experts. The HAQA corpus consists of 1598 question–answer pairs, and that of QUQA contains 3382. They may be useful as gold–standard datasets for the evaluation process, as training datasets for language models with question-answering tasks and for other uses in artificial intelligence.

The Qur’an QA 2023 shared task has two sub tasks: Passage Retrieval (PR) task and Machine Reading Comprehension (MRC) task. Our participation in the PR task was to further train several Arabic pre-trained models using a Sentence-Transformers architecture and to ensemble the best performing models. The results of the test set did not reflect the results of the development set. CL-AraBERT achieved the best results, with a 0.124 MAP. We also participate in the MRC task by further fine-tuning the base and large variants of AraBERT using Classical Arabic and Modern Standard Arabic datasets. Base AraBERT achieved the best result with the development set with a partial average precision (pAP) of 0.49, while it achieved 0.5 with the test set. In addition, we applied the ensemble approach of best performing models and post-processing steps to the final results. Our experiments with the development set showed that our proposed model achieved a 0.537 pAP. On the test set, our system obtained a pAP score of 0.49.

Question answering is a specialized area in the field of NLP that aims to extract the answer to a user question from a given text. Most studies in this area focus on the English language, while other languages, such as Arabic, are still in their early stage. Recently, research tend to develop question answering systems for Arabic Islamic texts, which may impose challenges due to Classical Arabic. In this paper, we use Simple Transformers Question Answering model with three Arabic pre-trained language models (AraBERT, CAMeL-BERT, ArabicBERT) for Qur’an Question Answering task using Qur’anic Reading Comprehension Dataset. The model is set to return five answers ranking from the best to worst based on their probability scores according to the task details. Our experiments with development set shows that AraBERT V0.2 model outperformed the other Arabic pre-trained models. Therefore, AraBERT V0.2 was chosen for the the test set and it performed fair results with 0.45 pRR score, 0.16 EM score and 0.42 F1 score.

Transformer-based models showed near-perfect results on several downstream tasks. However, their performance on classical Arabic texts is largely unexplored. To fill this gap, we evaluate monolingual, bilingual, and multilingual state-of-the-art models to detect relatedness between the Quran (Muslim holy book) and the Hadith (Prophet Muhammed teachings), which are complex classical Arabic texts with underlying meanings that require deep human understanding. To do this, we carefully built a dataset of Quran-verse and Hadith-teaching pairs by consulting sources of reputable religious experts. This study presents the methodology of creating the dataset, which we make available on our repository, and discusses the models’ performance that calls for the imminent need to explore avenues for improving the quality of these models to capture the semantics in such complex, low-resource texts.

The Quran, as a significant religious text, bears important spiritual and linguistic values. Understanding the text and inferring the underlying meanings entails semantic similarity analysis. We classified the verses of the Quran into 15 pre-defined categories or concepts, based on the Qurany corpus, using Doc2Vec and Logistic Regression. Our classifier scored 70% accuracy, and 60% F1-score using the distributed bag-of-words architecture. We then measured how similar the documents within the same category are to each other semantically and use this information to evaluate our model. We calculated the mean difference and average similarity values for each category to indicate how well our model describes that category.

Bidirectional Encoder Representations from Transformers (BERT) has gained popularity in recent years producing state-of-the-art performances across Natural Language Processing tasks. In this paper, we used AraBERT language model to classify pairs of verses provided by the QurSim dataset to either be semantically related or not. We have pre-processed The QurSim dataset and formed three datasets for comparisons. Also, we have used both versions of AraBERT, which are AraBERTv02 and AraBERTv2, to recognise which version performs the best with the given datasets. The best results was AraBERTv02 with 92% accuracy score using a dataset comprised of label ‘2’ and label '-1’, the latter was generated outside of QurSim dataset.

This article describes the process of gathering and constructing a bilingual parallel corpus of Islamic Hadith, which is the set of narratives reporting different aspects of the prophet Muhammad’s life. The corpus data is gathered from the six canonical Hadith collections using a custom segmentation tool that automatically segments and annotates the two Hadith components with 92% accuracy. This Hadith segmenter minimises the costs of language resource creation and produces consistent results independently from previous knowledge and experiences that usually influence human annotators. The corpus includes more than 10M tokens and will be freely available via the LREC repository.

In this paper we explore the use of Prediction by partial matching (PPM) compression based to segment Hadith into its two main components (Isnad and Matan). The experiments utilized the PPMD variant of the PPM, showing that PPMD is effective in Hadith segmentation. It was also tested on Hadith corpora of different structures. In the first experiment we used the non- authentic Hadith (NAH) corpus for train- ing models and testing, and in the second experiment we used the NAH corpus for training models and the Leeds University and King Saud University (LK) Hadith cor- pus for testing PPMD segmenter. PPMD of order 7 achieved an accuracy of 92.76% and 90.10% in the first and second experiments, respectively.

This is a pilot study that aims to explore the potential of using WEKA in forensic authorship analysis. It is a corpus-based research using data from Twitter collected from thirteen authors from Riyadh, Saudi Arabia. It examines the performance of unbalanced and balanced data sets using different classifiers and parameters of word grams. The attributes are dialect-specific linguistic features categorized as word grams. The findings further support previous studies in computational authorship identification.

This paper describes an Arabic dialect identification system which we developed for the Discriminating Similar Languages (DSL) 2016 shared task. We classified Arabic dialects by using Waikato Environment for Knowledge Analysis (WEKA) data analytic tool which contains many alternative filters and classifiers for machine learning. We experimented with several classifiers and the best accuracy was achieved using the Sequential Minimal Optimization (SMO) algorithm for training and testing process set to three different feature-sets for each testing process. Our approach achieved an accuracy equal to 42.85% which is considerably worse in comparison to the evaluation scores on the training set of 80-90% and with training set “60:40” percentage split which achieved accuracy around 50%. We observed that Buckwalter transcripts from the Saarland Automatic Speech Recognition (ASR) system are given without short vowels, though the Buckwalter system has notation for these. We elaborate such observations, describe our methods and analyse the training dataset.

This paper aims to implement what is referred to as the collocation of the Arabic keywords approach for extracting formulaic sequences (FSs) in the form of high frequency but semantically regular formulas that are not restricted to any syntactic construction or semantic domain. The study applies several distributional semantic models in order to automatically extract relevant FSs related to Arabic keywords. The data sets used in this experiment are rendered from a new developed corpus-based Arabic wordlist consisting of 5,189 lexical items which represent a variety of modern standard Arabic (MSA) genres and regions, the new wordlist being based on an overlapping frequency based on a comprehensive comparison of four large Arabic corpora with a total size of over 8 billion running words. Empirical n-best precision evaluation methods are used to determine the best association measures (AMs) for extracting high frequency and meaningful FSs. The gold standard reference FSs list was developed in previous studies and manually evaluated against well-established quantitative and qualitative criteria. The results demonstrate that the MI.log_f AM achieved the highest results in extracting significant FSs from the large MSA corpus, while the T-score association measure achieved the worst results.

Inspired by the Oxford Children’s Corpus, we have developed a prototype corpus of Arabic texts written and/or selected for children. Our Arabic Children’s Corpus of 2950 documents and nearly 2 million words has been collected manually from the web during a 3-month project. It is of high quality, and contains a range of different children’s genres based on sources located, including classic tales from The Arabian Nights, and popular fictional characters such as Goha. We anticipate that the current and subsequent versions of our corpus will lead to interesting studies in text classification, language use, and ideology in children’s texts.

In this paper, we focus on the prosodic effect of qalqalah or “vibration” applied to a subset of Arabic consonants under certain constraints during correct Qur’anic recitation or taǧwīd, using our Boundary-Annotated Qur’an dataset of 77430 words (Brierley et al 2012; Sawalha et al 2014). These qalqalah events are rule-governed and are signified orthographically in the Arabic script. Hence they can be given abstract definition in the form of regular expressions and thus located and collected automatically. High frequency qalqalah content words are also found to be statistically significant discriminators or keywords when comparing Meccan and Medinan chapters in the Qur’an using a state-of-the-art Visual Analytics toolkit: Semantic Pathways. Thus we hypothesise that qalqalah prosody is one way of highlighting salient items in the text. Finally, we implement Arabic transcription technology (Brierley et al under review; Sawalha et al forthcoming) to create a qalqalah pronunciation guide where each word is transcribed phonetically in IPA and mapped to its chapter-verse ID. This is funded research under the EPSRC “Working Together” theme.

This paper presents QurAna: a large corpus created from the original Quranic text, where personal pronouns are tagged with their antecedence. These antecedents are maintained as an ontological list of concepts, which have proved helpful for information retrieval tasks. QurAna is characterized by: (a) comparatively large number of pronouns tagged with antecedent information (over 24,500 pronouns), and (b) maintenance of an ontological concept list out of these antecedents. We have shown useful applications of this corpus. This corpus is first of its kind considering classical Arabic text, which could be used for interesting applications for Modern Standard Arabic as well. This corpus would benefit researchers in obtaining empirical and rules in building new anaphora resolution approaches. Also, such corpus would be used to train, optimize and evaluate existing approaches.

This paper presents a large corpus created from the original Quranic text, where semantically similar or related verses are linked together. This corpus will be a valuable evaluation resource for computational linguists investigating similarity and relatedness in short texts. Furthermore, this dataset can be used for evaluation of paraphrase analysis and machine translation tasks. Our dataset is characterised by: (1) superior quality of relatedness assignment; as we have incorporated relations marked by well-known domain experts, this dataset could thus be considered a gold standard corpus for various evaluation tasks, (2) the size of our dataset; over 7,600 pairs of related verses are collected from scholarly sources with several levels of degree of relatedness. This dataset could be extended to over 13,500 pairs of related verses observing the commutative property of strongly related pairs. This dataset was incorporated into online query pages where users can visualize for a given verse a network of all directly and indirectly related verses. Empirical experiments showed that only 33% of related pairs shared root words, emphasising the need to go beyond common lexical matching methods, and incorporate -in addition- semantic, domain knowledge, and other corpus-based approaches.

We train and test two probabilistic taggers for Arabic phrase break prediction on a purpose-built, “gold standard”, boundary-annotated and PoS-tagged Qur'an corpus of 77430 words and 8230 sentences. In a related LREC paper (Brierley et al., 2012), we cover dataset build. Here we report on comparative experiments with off-the-shelf N-gram and HMM taggers and coarse-grained feature sets for syntax and prosody, where the task is to predict boundary locations in an unseen test set stripped of boundary annotations by classifying words as breaks or non-breaks. The preponderance of non-breaks in the training data sets a challenging baseline success rate: 85.56%. However, we achieve significant gains in accuracy with the trigram tagger, and significant gains in performance recognition of minority class instances with both taggers via Balanced Classification Rate. This is initial work on a long-term research project to produce annotation schemes, language resources, algorithms, and applications for Classical and Modern Standard Arabic.

A boundary-annotated and part-of-speech tagged corpus is a prerequisite for developing phrase break classifiers. Boundary annotations in English speech corpora are descriptive, delimiting intonation units perceived by the listener. We take a novel approach to phrase break prediction for Arabic, deriving our prosodic annotation scheme from Tajwīd (recitation) mark-up in the Qur'an which we then interpret as additional text-based data for computational analysis. This mark-up is prescriptive, and signifies a widely-used recitation style, and one of seven original styles of transmission. Here we report on version 1.0 of our Boundary-Annotated Qur'an dataset of 77430 words and 8230 sentences, where each word is tagged with prosodic and syntactic information at two coarse-grained levels. In (Sawalha et al., 2012), we use the dataset in phrase break prediction experiments. This research is part of a larger-scale project to produce annotation schemes, language resources, algorithms, and applications for Classical and Modern Standard Arabic.

This paper describes the underlying software platform used to develop and publish annotations for the Quranic Arabic Corpus (QAC). The QAC (Dukes, Atwell and Habash, 2011) is a multimodal language resource that integrates deep tagging, interlinear translation, multiple speech recordings, visualization and collaborative analysis for the Classical Arabic language of the Quran. Available online at http://corpus.quran.com, the website is a popular study guide for Quranic Arabic, used by over 1.2 million visitors over the past year. We provide a description of the underlying software system that has been used to develop the corpus annotations. The multimodal data is made available online through an accessible cross-referenced web interface. Although our Linguistic Analysis Multimodal Platform (LAMP) has been applied to the Classical Arabic language of the Quran, we argue that our annotation model and software architecture may be of interest to other related corpus linguistics projects. Work related to LAMP includes recent efforts for annotating other Classical languages, such as Ancient Greek and Latin (Bamman, Mambrini and Crane, 2009), as well as commercial systems (e.g. Logos Bible study) that provide access to syntactic tagging for the Hebrew Bible and Greek New Testament (Brannan, 2011).

The Quranic Arabic Dependency Treebank (QADT) is part of the Quranic Arabic Corpus (http://corpus.quran.com), an online linguistic resource organized by the University of Leeds, and developed through online collaborative annotation. The website has become a popular study resource for Arabic and the Quran, and is now used by over 1,500 researchers and students daily. This paper presents the treebank, explains the choice of syntactic representation, and highlights key parts of the annotation guidelines. The text being analyzed is the Quran, the central religious book of Islam, written in classical Quranic Arabic (c. 600 CE). To date, all 77,430 words of the Quran have a manually verified morphological analysis, and syntactic analysis is in progress. 11,000 words of Quranic Arabic have been syntactically annotated as part of a gold standard treebank. Annotation guidelines are especially important to promote consistency for a corpus which is being developed through online collaboration, since often many people will participate from different backgrounds and with different levels of linguistic expertise. The treebank is available online for collaborative correction to improve accuracy, with suggestions reviewed by expert Arabic linguists, and compared against existing published books of Quranic Syntax.

Morphological analyzers and part-of-speech taggers are key technologies for most text analysis applications. Our aim is to develop a part-of-speech tagger for annotating a wide range of Arabic text formats, domains and genres including both vowelized and non-vowelized text. Enriching the text with linguistic analysis will maximize the potential for corpus re-use in a wide range of applications. We foresee the advantage of enriching the text with part-of-speech tags of very fine-grained grammatical distinctions, which reflect expert interest in syntax and morphology, but not specific needs of end-users, because end-user applications are not known in advance. In this paper we review existing Arabic Part-of-Speech Taggers and tag-sets, and illustrate four different Arabic PoS tag-sets for a sample of Arabic text from the Quran. We describe the detailed fine-grained morphological feature tag set of Arabic, and the fine-grained Arabic morphological analyzer algorithm. We faced practical challenges in applying the morphological analyzer to the 100-million-word Web Arabic Corpus: we had to port the software to the National Grid Service, adapt the analyser to cope with spelling variations and errors, and utilise a Broad-Coverage Lexical Resource combining 23 traditional Arabic lexicons. Finally we outline the construction of a Gold Standard for comparative evaluation.

Broad-coverage language resources which provide prior linguistic knowledge must improve the accuracy and the performance of NLP applications. We are constructing a broad-coverage lexical resource to improve the accuracy of morphological analyzers and part-of-speech taggers of Arabic text. Over the past 1200 years, many different kinds of Arabic language lexicons were constructed; these lexicons are different in ordering, size and aim or goal of construction. We collected 23 machine-readable lexicons, which are freely available on the web. We combined lexical resources into one large broad-coverage lexical resource by extracting information from disparate formats and merging traditional Arabic lexicons. To evaluate the broad-coverage lexical resource we computed coverage over the Qur’an, the Corpus of Contemporary Arabic, and a sample from the Arabic Web Corpus, using two methods. Counting exact word matches between test corpora and lexicon scored about 65-68%; Arabic has a rich morphology with many combinations of roots, affixes and clitics, so about a third of words in the corpora did not have an exact match in the lexicon. The second approach is to compute coverage in terms of use in a lemmatizer program, which strips clitics to look for a match for the underlying lexeme; this scored about 82-85%.

We have previously reported on ProPOSEL, a purpose-built Prosody and PoS English Lexicon compatible with the Python Natural Language ToolKit. ProPOSEC is a new corpus research resource built using this lexicon, intended for distribution with the Aix-MARSEC dataset. ProPOSEC comprises multi-level parallel annotations, juxtaposing prosodic and syntactic information from different versions of the Spoken English Corpus, with canonical dictionary forms, in a query format optimized for Perl, Python, and text processing programs. The order and content of fields in the text file is as follows: (1) Aix-MARSEC file number; (2) word; (3) LOB PoS-tag; (4) C5 PoS-tag; (5) Aix SAM-PA phonetic transcription; (6) SAM-PA phonetic transcription from ProPOSEL; (7) syllable count; (8) lexical stress pattern; (9) default content or function word tag; (10) DISC stressed and syllabified phonetic transcription; (11) alternative DISC representation, incorporating lexical stress pattern; (12) nested arrays of phonemes and tonic stress marks from Aix. As an experimental dataset, ProPOSEC can be used to study correlations between these annotation tiers, where significant findings are then expressed as additional features for phrasing models integral to Text-to-Speech and Speech Recognition. As a training set, ProPOSEC can be used for machine learning tasks in Information Retrieval and Speech Understanding systems.

This paper describes experimental use of the multi-agent architecture to integrate Natural Language and Information Systems research and teaching, by casting a group of students as intelligent agents to collect and analyse English language resources from around the world. Section 2 and section 3 describe the hybrid intelligent information systems experiments at the University of Leeds and the results generated, including several research papers accepted at international conferences, and a finalist entry in the British Computer Society Machine Intelligence contest. Our proposals for applying the multi-agent idea in other universities such as the Arab Open University are presented in section 4. The conclusion is presented in section 5: the success of hybrid intelligent information systems experiments in generating research papers within a limited time.

ProPOSEL is a prototype prosody and PoS (part-of-speech) English lexicon for Language Engineering, derived from the following language resources: the computer-usable dictionary CUVPlus, the CELEX-2 database, the Carnegie-Mellon Pronouncing Dictionary, and the BNC, LOB and Penn Treebank PoS-tagged corpora. The lexicon is designed for the target application of prosodic phrase break prediction but is also relevant to other machine learning and language engineering tasks. It supplements the existing record structure for wordform entries in CUVPlus with syntactic annotations from rival PoS-tagging schemes, mapped to fields for default closed and open-class word categories and for lexical stress patterns representing the rhythmic structure of wordforms and interpreted as potential new text-based features for automatic phrase break classifiers. The current version of the lexicon comes as a textfile of 104052 separate entries and is intended for distribution with the Natural Language ToolKit; it is therefore accompanied by supporting Python software for manipulating the data so that it can be used for Natural Language Processing (NLP) and corpus-based research in speech synthesis and speech recognition.

Ce papier discute la Néoposie: l’inférence auto-adaptive de catégories grammaticales de mots de la langue naturelle. L’inférence grammaticale peut être divisée en deux parties : l’inférence de catégories grammaticales de mots et l’inférence de la structure. Nous examinons les éléments de base de l’apprentissage auto-adaptif du marquage des catégories grammaticales, et discutons l’adaptation des trois types principaux de marqueurs des catégories grammaticales à l’inférence auto-adaptive de catégories grammaticales de mots. Des marqueurs statistiques de n-grammes suggèrent une approche de regroupement statistique, mais le regroupement n’aide ni avec les types de mots peu fréquents, ni avec les types de mots nombreux qui peuvent se présenter dans plus d’une catégorie grammaticale. Le marqueur alternatif d’apprentissage basé sur la transformation suggère une approche basée sur la contrainte de l’unification de contextes d’occurrences de mots. Celle-ci présente un moyen de regrouper des mots peu fréquents, et permet aux occurrences différentes d’un seul type de mot d’appartenir à des catégories différentes selon les contextes grammaticaux où ils se présentent. Cependant, la simple unification de contextes d’occurrences de mots produit un nombre incroyablement grand de catégories grammaticales de mots. Nous avons essayé d’unifier plus de catégories en modérant le contexte de la correspondance pour permettre l’unification des catégories de mots aussi bien que des occurrences de mots, mais cela entraîne des unifications fausses. Nous concluons que l’avenir peut être un hybride qui comprend le regroupement de types de mots peu fréquents, l’unification de contextes d’occurrences de mots, et le ‘seeding’ avec une connaissance linguistique limitée. Nous demandons un programme de nouvelles recherches pour développer une valise pour la découverte de la langue naturelle.

Existing automated MT evaluation methods often require expert human translations. These are produced for every language pair evaluated and, due to this expense, subsequent evaluations tend to rely on the same texts, which do not necessarily reflect real MT use. In contrast, we are designing an automated MT evaluation system, intended for use by post-editors, purchasers and developers, that requires nothing but the raw MT output. Furthermore, our research is based on texts that reflect corporate use of MT. This paper describes our first step in system design: a hierarchical classification scheme of fluency errors in English MT output, to enable us to identify error types and frequencies, and guide the selection of errors for automated detection. We present results from the statistical analysis of 20,000 words of MT output, manually annotated using our classification scheme, and describe correlations between error frequencies and human scores for fluency and adequacy.