Even with various regulations in place across countries and social media platforms (Government of India, 2021; European Parliament and Council of the European Union, 2022), digital abusive speech remains a significant issue. One potential approach to address this challenge is automatic text detoxification, a text style transfer (TST) approach that transforms toxic language into a more neutral or non-toxic form. To date, the availability of parallel corpora for the text detoxification task (Logacheva et al., 2022; Atwell et al., 2022; Dementieva et al., 2024a) has proven to be crucial for state-of-the-art approaches. With this work, we extend parallel text detoxification corpus to new languages—German, Chinese, Arabic, Hindi, and Amharic—testing in the extensive multilingual setup TST baselines. Next, we conduct the first of its kind an automated, explainable analysis of the descriptive features of both toxic and non-toxic sentences, diving deeply into the nuances, similarities, and differences of toxicity and detoxification across 9 languages. Finally, based on the obtained insights, we experiment with a novel text detoxification method inspired by the Chain-of-Thoughts reasoning approach, enhancing the prompting process through clustering on relevant descriptive attributes.

Abstract Modelling the distributional semantics of such a morphologically rich language as Arabic needs to take into account its introflexive, fusional, and inflectional nature attributes that make up its combinatorial sequences and substitutional paradigms. To evaluate such word distributional models, the benchmarks that have been used thus far in Arabic have mimicked those in English. This paper reports on a benchmark that we designed to reflect linguistic patterns in both Contemporary Arabic and Classical Arabic, the first being a cover term for written and spoken Modern Standard Arabic, while the second for pre-modern Arabic. The analogy items we included in this benchmark are chosen in a transparent manner such that they would capture the major features of nouns and verbs; derivational and inflectional morphology; high-, middle-, and low-frequency patterns and lexical items; and morphosemantic, morphosyntactic, and semantic dimensions of the language. All categories included in this benchmark are carefully selected to ensure proper representation of the language. The benchmark consists of 45 roots of the trilateral, all-consonantal, and semivowel-inclusive types; six morphosemantic patterns (’af‘ala; ifta‘ala; infa‘ala; istaf‘ala; tafa‘‘ala; and tafā‘ala); five derivations (the verbal noun, active participle, and the contrasts in Masculine-Feminine; Feminine-Singular-Plural; Masculine-Singular-Plural); and morphosyntactic transformations (perfect and imperfect verbs conjugated for all pronouns); and lexical semantics (synonyms, antonyms, and hyponyms of nouns, verbs, and adjectives), as well as capital cities and currencies. All categories include an equal proportion of high-, medium-, and low-frequency items. For the purpose of validating the proposed benchmark, we developed a set of embedding models from different textual sources. Then, we tested them intrinsically using the proposed benchmark and extrinsically using two natural language processing tasks: Arabic Named Entity Recognition and Text Classification. The evaluation leads to the conclusion that the proposed benchmark is truly reflective of this morphologically rich language and discriminatory of word embeddings.

This work presents a detailed comparison of the performance of deep learning models such as convolutional neural networks, long short-term memory, gated recurrent units, their hybrids, and a selection of shallow learning classifiers for sentiment analysis of Arabic reviews. Additionally, the comparison includes state-of-the-art models such as the transformer architecture and the araBERT pre-trained model. The datasets used in this study are multi-dialect Arabic hotel and book review datasets, which are some of the largest publicly available datasets for Arabic reviews. Results showed deep learning outperforming shallow learning for binary and multi-label classification, in contrast with the results of similar work reported in the literature. This discrepancy in outcome was caused by dataset size as we found it to be proportional to the performance of deep learning models. The performance of deep and shallow learning techniques was analyzed in terms of accuracy and F1 score. The best performing shallow learning technique was Random Forest followed by Decision Tree, and AdaBoost. The deep learning models performed similarly using a default embedding layer, while the transformer model performed best when augmented with araBERT.

Imbalanced data occurs in a wide range of scenarios. The skewed distribution of the target variable elicits bias in machine learning algorithms. One of the popular methods to combat imbalanced data is to artificially balance the data through resampling. In this paper, we compare the efficacy of a recently proposed kernel density estimation (KDE) sampling technique in the context of artificial neural networks. We benchmark the KDE sampling method against two base sampling techniques and perform comparative experiments using 8 datasets and 3 neural networks architectures. The results show that KDE sampling produces the best performance on 6 out of 8 datasets. However, it must be used with caution on image datasets. We conclude that KDE sampling is capable of significantly improving the performance of neural networks.

This paper proposes a CNN-based deep learning model that classifies Arabic poems based on its era, which is not reported before. To build this model, constructing a dataset is the first step, so we propose an updated Arabic Poetry Dataset (2020). We use FastText word embeddings, based on the full corpus of poems (unlabeled). Two classifiers were trained, namely, a supervised deep learning classifier and a FastText-based classifier. We conducted several experiments. First, we implemented a polarity classifier of poems to modern and non-modern eras, which achieved highest accuracy and F1-score of 0.913 and 0.914, respectively, using a deep learning model without frequent terms. In the second experiment, we categorized poems into three eras. The classifier reported an accuracy and F1-score of 0.875 each. Last, the classification of poems into five different eras achieved highest accuracy and F1-score of 0.801 and 0.796, respectively.

Question pairing is the task of listing similar question-answer pairs to a query question automatically. This technique is used by famous online question-answering forums like Quora and Stack exchange, and has gotten a lot of attention lately as it reduces the amount of duplicate questions on such topics. That is why there are many researchers working on this task on the English language, but there is not much focus on other languages like the Arabic language. This paper introduces a deep neural networks based solution to the question pairing task on Arabic questions using minimal pre-processing. We investigate the best settings for the proposed DNN model. We show a thorough set of experiments and demonstrate that our proposed model outperforms the few existing implementations reported in literature on benchmark datasets by no less than 10% in prediction accuracy measure. Namely, the NSURL 2019 question-question dataset and the SemEval 2017 question-answer dataset. Our proposed model achieved 77% and 58% prediction accuracy on both datasets, respectively.

This paper describes a supervised classification system of Quranic audio clips of several reciters. The objective is to identify the reciter or the closest reciter to an input audio clip. It is common to find people who like to recite Quran mimicking one of the popular reciters. To achieve a practical classifier system, we constructed a representative dataset of audio clips for seven popular reciters from Saudi Arabia. Key features were extracted from the audio clips. We chose perceptual features such as pitch and tempo based features, short time energy etc. We have tried different combination of perceptual features in order to achieve better classification. We split the dataset into training and testing sets (80% & 20%, respectively). SVM is used to implement the classifier. Experimental results show that the proposed audio classifier produces promising results with an overall accuracy of 90%.

The majority of available sentiment analysis systems treat English text, which resulted in a significant number of resources to assist researchers in this field. Unfortunately, this is not the case for other languages and, in particular, for the Arabic language. Few serious research attempts are reported to build necessary resources such as datasets to serve Arabic sentiment analysis research efforts. In order to expedite building necessary huge datasets in Arabic or any other dataset-poor language, automatic translation from a rich source-language can make a viable alternative to generate the required resources for a target language. This is mainly driven by the assumption that translation usually preserves the sentiment upto certain degree. In this work, we study the viability of this approach to automatically translate a huge dataset from English to Arabic for sentiment analysis purposes. Experimental results show that sentiment analysis of both original (English) dataset and the translated one (Arabic) produce comparable results. As a result, automatically translated English datasets shall add to the required resources for building robust Arabic sentiment analysis systems. Although automated translation may yield poor translations in terms of readability and comprehension by humans, sentiment analysis systems can still predict the majority of correct sentiments. Such sentiment is preserved by key words. We believe automated translation may be considered as a viable option to produce rich and well represented datasets in order to develop robust and efficient Arabic sentiment analyzers.

Many natural language processing tasks depend on the quality of word representations in order to achieve good results. To build good word vectors for Arabic sentiment analysis there are various factors to consider, such as the corpus choice, text preprocessing, and the choice of the training model. In this paper, we demonstrate how to achieve better sentiment analysis results by building sentiment-specific embeddings trained using the unsupervised fastText tool for both CBOW and skip-gram models. The results show that fastText embedding models are a valuable alternative to recover semantic and syntactic information of standard Arabic as well as a dialectal language. Indeed, fastText is more relevant to morphologically rich languages such as Arabic language. In order to further validate the effectiveness of the resulting word embeddings, we implemented polarity sentiment classifiers and compared their performance to 2 similar models reported in the literature on three different datasets. The performance evaluation results confirm that our approach outperforms counterpart ones on relevant datasets with an improvement margin of up to 5% using F1-score.

Arabic language has very rich vocabulary. It is manifested in different forms. The formal, Modern Standard Arabic (MSA), and the informal, colloquial or dialects. Dialectical languages become important as a result of the proliferation of social networks which resulted in the vast unstructured dialectical text available on the web. Unique properties of modern standard Arabic and dialects present major challenges to build sentiment analysis systems by adopting similar models designed for the English language. In this paper, we present a supervised Arabic sentiment analysis using a bag-of-words feature. We further examine using a set of key words (lexicon) for better polarity classification. The testing of the system is carried out on the freely-available Arabic books' reviews (LABR) dataset. LABR includes both modern standard Arabic and Egyptian dialectal reviews. We used both balanced and unbalanced datasets. Clearly, the balanced data set is small in size and, henceforth, a large-scale balanced dataset is required for training of the classifier model. Further, we compared the computed predicted sentiments against the actual reviews for a specific book. Findings, by annotators, had indicated ambiguity between a review and its rating when verified alongside the predicted sentiment, which provided a more reasonable result. Moreover, working with dialects and sarcasm is exceedingly exciting. Experimental results on the adopted logistic classifier model and LABR are encouraging and promising. However, a key prerequisite is the availability of rich and well represented datasets in order to develop robust and efficient Arabic sentiment analyzers.

The availability of rich datasets is a pre-requisite for proposing robust sentiment analysis systems. A variety of such datasets exists in English language. However, it is rare or nonexistent for the Arabic language except for a recent LABR dataset, which consists of a little bit over 63,000 book reviews extracted from. Goodreads. com. We introduce BRAD 1.0, the largest Book Reviews in Arabic Dataset for sentiment analysis and machine language applications. BRAD comprises of almost 510,600 book records. Each record corresponds for a single review and has the review in Arabic language and the reviewer's rating on a scale of 1 to 5 stars. In this paper, we present and describe the properties of BRAD. Further, we provide two versions of BRAD: the complete unbalanced dataset and the balanced version of BRAD. Finally, we implement four sentiment analysis classifiers based on this dataset and report our findings. When training and testing the classifiers on BRAD as opposed to LABR, an improvement rate growth of 46% is reported. The highest accuracy attained is 91%. Our core contribution is to make this benchmark-dataset available and accessible to the research community on Arabic language.

This paper presents a novel approach for the problem of tracking a moving target in a global dynamic environment. The robot has to move such that it keeps the target visible for the longest time possible, and at the same time, avoid colliding with any of the moving obstacles. This paper presents a solution that is based on the idea of three interacting components which perform: tracking, collision avoidance and motion selection. The proposed solution is validated using a comprehensive set of simulations, which show that transition from tracking in static environments to tracking in dynamic environments can be done without much loss in robot safety or tracking ability.

Learning how to program is a universal problem that is facing many students in introductory programming courses. This multinational problem created the need for an effective and easy to use learning system. The system introduced in this paper, Java Learning System using Dependence Graphs (JLearn-DG), teaches students the basic concepts of programming by creating the System Dependence Graph (SDG) for students program. The system defines four types of dependencies: control, method call, method library and data. It identifies the dependencies and relationships among all statements. When JLean-DG is compared to existing systems, it has two advantages: the ability to create the SDG in the presence of semantic-errors and the ability to create the dependencies among all statements in the program regardless of the execution. Based on our experimental evaluation, this tool has been an affective one and has good potential to teach programming.

This paper presents a practical algorithm for evader detection and tracking using one or more pursuers. The solution employs two advanced data structures. The first one is the rapidly-exploring random tree (RRT). It is constructed randomly but evenly distributed to generate a roadmap that captures the connectivity of the free space. The second data structure is the k-dimensional tree (Kd-Tree). Upon completion of the RRTs construction, their vertices are inserted in a Kd-Tree. At the tracking phase, the Kd-Tree will be queried repeatedly for retrieving the set of potential locations to be used by each pursuer in order to monitor or track an evader. Thus, the usage of the Kd-Tree will reduce the querying cost, during tracking, to a logarithmic time. A lazy collision detection strategy is used to resolve collisions with obstacles at runtime. As a result unnecessary checks are eliminated and hence improving the system performance. Simulation results show the validity of the proposed algorithm.

In this paper, we present a novel art gallery-based algorithm for placing a small number of guards to cover a global environment with obstacles for an autonomous robot. The guards are required to cover the entire workspace that is represented as a simple polygon with n vertices and h holes (obstacles). The proposed algorithm efficiently computes a (small) number of guards in simple polygons with holes, which runs in O (n log n) time and requires a linear storage complexity. However, an additional set of connection nodes is computed to form the connectivity graph, which contains all guards. This graph has far less number of vertices when compared to similar data structures used in conventional visibility-based or probabilistic-based motion planning algorithms. The resulting placement of guards and connectors can then be used as control points in the path of an autonomous mobile robot for navigation or inspection tasks. The proposed algorithm is not only offering a better performance in terms of computational cost but also an easy implementation. Simulation results demonstrate the efficiency, robustness, and potential of the proposed algorithm.

The aim of this paper is to evaluate the performance of our proposed art gallery-based roadmap algorithm against well-known and frequently cited visibility-based motion planning algorithms: the visibility graph and the visibility-based probabilistic roadmap. The comparison involves several criteria among which are: the cardinality of the graph, the completeness of the algorithm, coverage and connectivity of the free configuration space (CS/sub free/) and computational cost. Our proposed algorithm is robust and fast as it generally covers the whole CS/sub free/, based on the well-known art gallery theorem. It efficiently seeks to construct a roadmap that contains the smallest possible number of nodes (called guards) as opposed to generating a large number of nodes when compared to other motion planning approaches. The simulation results demonstrate that our proposed algorithm outperforms both algorithms and proves to not only combine the attractive features of both algorithms but also eliminate the drawbacks of each one.

This paper addresses an important and vital problem within the general area of character recognition, namely separating connected digits in handwritten numerals. The basic idea is employing multiagents to handle this problem. Our approach is mainly based on detecting the deepest-top valley and the highest-bottom hill, with one agent dedicated to each. The former agent decides on candidate cut-point as the closest feature point to the center of the deepest top-valley, if any. On the other hand, the other agent argues candidate cut-point as the closest feature point to the center of the highest bottom-hill, if any. After each of the two agents applies its own rules and reports a candidate cut-point, the two agents negotiate to agree on the actual cut-point. It may happen that both agents come to the negotiation with the same candidate cut-point. However, in general, each agent may find a different candidate cut-point and they negotiate a compromise on the actual cut-point, which may be neither of the two already proposed cut-points. A degree of confidence in each candidate cut-point influences the negotiation process. Experiments conducted so far are promising and successful. The obtained results are very encouraging with a success factor of 97.8%. Finally, none of the two agents alone achieved a close success rate.

This paper presents a novel approach to separateconnected digits in handwritten numerals by employing twoagents in the process. The first agent decides oncandidate cut-point as the closest feature-point to the center ofthe deepest top-valley, if any. The second agent arguescandidate cut-point as the closest feature-point to the center ofthe highest bottom-hill, if any. Then the actual cut-point isdecided by negotiation, which is influenced by a degree ofconfidence in each candidate cut-point. Experimentscon-ducted so far are promising and successful as well asjustified employing multiple agents. The obtained results arevery encouraging with a success factor of 97.8%.

In this paper, we describe an algorithm for predicting future positions and orientation of a moving object in a time-varying environment using an autoregressive model (ARM). No constraint is placed on the obstacles motion. The model addresses prediction of translational and rotational motions. Rotational motion is represented using quaternions rather than Euler representation to improve the algorithm performance and accuracy of the prediction results. Compared to other similar systems, the proposed algorithm has an adaptive capability, which enables it to predict over multiple time-steps rather than fixed ones as reported in other works. Such algorithm can be used in a variety of applications. An important one is its application in the framework of designing reliable navigational systems for autonomous mobile robots and more particularly in building effective trajectory planners. Simulation results show how significantly this model could reduce computational cost.

This paper presents a new algorithm to the motion planning problem for a mobile robot in a local environment. The algorithm uses the step-by-step approach to move the robot along the line that joins the starting point and the goal point. When an obstacle is met the algorithm assigns a value to each one of the eight surrounding points that surround the current point then the robot moves to the point that has the minimum value. Each obstacle point is assigned a relatively large value, while the value assigned to a non-obstacle point is the algebraic sum of the horizontal and the vertical distances from the goal point. The proposed algorithm completely eliminates the local minima problem, which is exhibited in most artificial-potential based methods, and most important it eliminates the "flat-regions" problem. The proposed algorithm is superior to other algorithms because of its low-computational cost besides simplicity. It also outperforms the blind-follow based algorithms. Simulation results demonstrates the validity and the potential of the proposed algorithm.

A method for recognizing handwritten Hindi numerals is proposed based on the structural descriptors of a numeral's shape. The method consists of three major steps. The first one is preprocessing, where a handwritten numeral is scanned, normalized and then thinned. Next, a robust algorithm is used to segment the scanned image into stroke(s), based on feature points, and to identify cavity features. The output of this algorithm is a syntactic representation (that is one or more syntactic terms). Finally, this syntacytic representation is matched against the set of prototype syntactic representations of handwritten numerals for a possible match. Early experimental results are not only encouraging but also proving the tolerance of the proposed system to recognize a high variability of Hindi numerals' shapes. The system attained a successful recognition rate of 96%.

This paper presents a new formulation of the artificial potential approach to the motion planning problem for a mobile robot in a global environment. To model the potential (magnetic) field by Maxwell's Equations that completely eliminate the local minima problem, which is exhibited in most artificial potential methods, such as Harmonic functions based methods. However, the proposed model is superior to the Harmonic one because it is easily extendable to 3D, the time dimension is modeled by default which means it is a suitable model for time-varying environments, computations are less and hence faster, and most important it eliminates the "flat-regions" problem. In this work, electrical currents are assumed to be floating in a cluttered environment with obstacles. The obstacles are assigned zero conductivity whereas the goal point is assigned the highest electrical conductivity. The magnetic field induced by the electric currents is used to find a free path between the start and goal points. Simulation results reflects the validity and the potential of the proposed model.

In this paper the theoretical basis is presented and the implementation of a term rewriting system based on algebraic specifications is described. The input to this system is represented by an algebraic specification language, which forms not only the set of axioms but also the sorts, variables, operators and terms of a specific simulated theory or application. Rewriting and matching mechanisms provide the formal methodology for evaluating terms and proving assertions in an algebraic theory. Specifications are evaluated by interpreting terms by means of rewrite rules. The rules are described by the axioms of the specifications where the finite termination and congruence properties are assumed. A term rewriting system to recognize handwritten Hindu numerals is introduced as a case study. Besides rewriting, a robust algorithm is proposed to segment the numeral's image into strokes based on feature points and to identify cavity features. A syntactic representation (term) of the input image is matched and rewritten against a set of rules. Experimental results proved that the proposed system is tolerant to recognize a variety of numeral shapes with 96% successful recognition rate.

The paper addresses an important and vital problem within the general area of character recognition, namely the identification and recognition of touching in handwritten Hindi numerals. The basic idea is that while writing down numbers, it is possible to have adjacent digits touching each other. To handle this, we are developing a multi-agent system. So far, we have two agents, which are presented. The first agent works directly on the scanned image of the original handwritten number. It locates possible touching based on the thickness of handwriting. The other works on the thinned image. It segments the image into four categories of segments and tries to locate possible touching based on the rules that govern the connection of segments to form digits. After each of the two agents applies its own rules and investigates possible touching, and to increase touching recognition rate, the two agents negotiate and try to agree on the actual touching. The experiments carried out so far are promising and successful. The obtained results are very encouraging with a success factor of 92.70%.

An approach to generating acceleration-based optimal smooth piecewise trajectories is proposed. Given two configurations (position and orientation) in 3D, we search for the minimal energy trajectory that minimizes the integral of the squared acceleration, opposed to curvature, which is widely investigated. The variation in both components of acceleration controls the smoothness of generated trajectories. Our objective is to search for the trajectory along which a free moving robot is able to accelerate (decelerate) to a safe speed in an optimal way. A numerical iterative procedure is devised for computing the optimal piecewise trajectory as a solution of a constrained boundary value problem.

In this paper, we describe a framework for predicting future positions and orientation of moving obstacles in a time-varying environment using autoregressive model (ARM) with conditional maximum likelihood estimate of the model parameters. No constraints are placed on the obstacles motion. The proposed algorithm can be used in a variety of applications, one of which is robot motion planning in time varying environments.

A novel approach to generating optimal smooth piecewise trajectories based on a new energy measure is proposed. Given the configurations (position and direction) of two points in the plane, we search for the minimal energy trajectory that minimizes the integral of the squared acceleration opposed to curvature, which has been the predominant energy measure studied in the literature. The smoothness of the optimal trajectory depends on how the tangential and normal components of acceleration vary over an interval of time. A numerical iterative procedure is devised for computing the optimal piecewise trajectory as a solution of a constrained boundary value problem. The resulting trajectories are not only smooth but also safe with optimal velocity (acceleration) profiles and therefore suitable for robot motion planning applications. The feasibility of the proposed approach is illustrated by several simulation examples. Besides motion planning, the resulting trajectories may be useful in computer graphics and geometric design.

In underwater environments it is often difficult to obtain a big/clear picture of a scene. For that reason, a system that can integrate small pieces of images (taken from close range) into a composite 3D surface, is developed here. The device, along with 3D position/orientation estimation equipment, can be used for inspection of hulls of ships anchored in a bay, or for examination of underwater pipes and tanks. Experimental results are presented which validate the algorithms developed.

We introduce a technique for detecting moving objects from an image sequence obtained with a moving camera using the planarity constraint. To increase the robustness of this technique, false motion caused by inaccuracies in sensor readings is eliminated by use of a morphological filter. This involves two successive operations-erosion and dilation-performed on a motion compensated image. Experimental results with real images are presented. Applications to the compression of moving images are now being investigated.

An approach to generating smooth piecewise local trajectories for mobile robots is proposed. Given the configurations (position and direction) of two points, one searches for the trajectory that minimizes the integral of acceleration (tangential and normal). The resulting trajectory should not only be smooth but also safe in order to be applicable in real-life situations, so the authors investigate two different obstacle-avoidance constraints that satisfy the minimization problem. Unfortunately, in this case the problem becomes more complex and unsuitable for real-time implementations. Therefore, the authors introduce two simple solutions, based on the idea of polynomial fitting to generate safe trajectories, once a collision is detected with the original smooth trajectory. Simulation results for the different algorithms are presented.

A heuristic technique for solving the problem of path planning based on local information for a mobile robot with acceleration constraints moving amidst a set of stationary obstacles is described. The concept of safety is used to design a planning strategy. A path based on local information that maximizes the product of safety and attraction towards the goal is chosen. The safety function depends on the acceleration bounds. The attraction towards the goal depends on the distance from the goal. Two additional heuristics are proposed to improve the efficiency of the search process, and to enhance the ability of the robot to avoid obstacles.>

The authors describe a heuristic technique for solving the problem of path planning based on local information for a mobile robot with acceleration constraints moving amidst a set of stationary obstacles. The concept of safety is introduced to design a planning strategy. A path which maximizes the product of safety based on local information and attraction towards the goal is chosen. The safety function depends on the acceleration bounds. The attraction toward the goal depends on the distance from the goal. Two additional heuristics are proposed to improve the efficiency of the search process and to enhance the ability of the robot to avoid obstacles. Some simulation examples of the algorithm corresponding to different navigational environments are discussed.>