Text Alignment Research Articles

A multiscale feature fusion method for cursive text detection in natural scene images

ABSTRACT Text detection in natural images is a challenging problem due to variations in text size, aspect ratio, alignment and background complexity. This paper proposes a multiscale feature fusion convolutional neural network method to detect cursive and multi-language text in natural images. The proposed method combines VGG-16 features at multi-scales and multi-layers and creates a new convolutional feature map of shallow and deep layers. On top of convolutional feature map, a vertical text proposal generation method is used that generates fixed-size text proposals. A recurrent layer is implemented which takes the convolutional feature maps of window as sequential input and updates the recurrent state internally in the hidden layers. The output of recurrent layer is mapped to the two fully connected layers to predict the text/non-text region proposals and bounding boxes regression. The model is evaluated on a custom-developed Urdu scene text dataset and the ICDAR-MLT17 Arabic text image dataset.

Inscriptis - A Python-based HTML to text conversion library optimized for knowledge extraction from the Web

Inscriptis provides a library, command line client and Web service for converting HTML to plain text. Its development has been triggered by the need to obtain accurate text representations for knowledge extraction tasks that preserve the spatial alignment of text without drawing upon heavyweight, browser-based solutions such as Selenium. In contrast to related software packages, Inscriptis (i) provides a layout-aware conversion of HTML that more closely resembles the rendering obtained from standard Web browsers; and (ii) supports annotation rules, i.e., user-provided mappings that allow for annotating the extracted text based on structural and semantic information encoded in HTML tags and attributes. These unique features ensure that downstream knowledge extraction components can operate on accurate text representations, and may even use information on the semantics and structure of the original HTML document.

Fine-Grained Cross-Modal Retrieval for Cultural Items with Focal Attention and Hierarchical Encodings

In this paper, we target the tasks of fine-grained image–text alignment and cross-modal retrieval in the cultural heritage domain as follows: (1) given an image fragment of an artwork, we retrieve the noun phrases that describe it; (2) given a noun phrase artifact attribute, we retrieve the corresponding image fragment it specifies. To this end, we propose a weakly supervised alignment model where the correspondence between the input training visual and textual fragments is not known but their corresponding units that refer to the same artwork are treated as a positive pair. The model exploits the latent alignment between fragments across modalities using attention mechanisms by first projecting them into a shared common semantic space; the model is then trained by increasing the image–text similarity of the positive pair in the common space. During this process, we encode the inputs of our model with hierarchical encodings and remove irrelevant fragments with different indicator functions. We also study techniques to augment the limited training data with synthetic relevant textual fragments and transformed image fragments. The model is later fine-tuned by a limited set of small-scale image–text fragment pairs. We rank the test image fragments and noun phrases by their intermodal similarity in the learned common space. Extensive experiments demonstrate that our proposed models outperform two state-of-the-art methods adapted to fine-grained cross-modal retrieval of cultural items for two benchmark datasets.

Real-Time Arabic Text-Reading for Visually Impaired People

Visually impaired (VI) people suffer from many difficulties when accessing printed material using existing technologies. These problems may include text alignment, focus, accuracy, software processing speed, mobility, and efficiency. Current technologies such as flatbed scanners and OCR programs need to scan an entire page. Recently, VI people prefer mobile devices because of their handiness and accessibility, but they have problems with focusing the mobile camera on the printed material. In this paper, a real-time Arabic text-reading prototype for VI people is proposed. It is based on using a wearable device for a hand finger. It is designed as a wearable ring attached to a tiny webcam device. The attached camera captures the printed Arabic text and passes it to the Arabic OCR system. Finally, the recognized characters are translated into speech using the text-to-speech (TTS) technology. Experimental results demonstrate the feasibility of the proposed prototype. It achieved an accuracy of 95.86% for Arabic character recognition and 98.5% for English character recognition.

A Survey of Text Alignment Visualization.

Text alignment is one of the fundamental techniques text-related domains like natural language processing, computational linguistics, and digital humanities. It compares two or more texts with each other aiming to find similar textual patterns, or to estimate in general how different or similar the texts are. Visualizing alignment results is an essential task, because it helps researchers getting a comprehensive overview of individual findings and the overall pattern structure. Different approaches have been developed to visualize and help making sense of these patterns depending on text size, alignment methods, and, most importantly, the underlying research tasks demanding for alignment. On the basis of those tasks, we reviewed existing text alignment visualization approaches, and discuss their advantages and drawbacks. We finally derive design implications and shed light on related future challenges.

Laptops in the Auditorium: Facing Educational Challenges in Classics by Teaching Digital Tools

Higher education in the Arts and Humanities nowadays faces global challenges. By its very nature, Classics is among the disciplines which experience the harshest consequences of the large economic and cultural shifts we are witnessing. In the last decades, the students graduating from the BA curriculum of the Department of Classics to the University of Sofia have decreased in number, their profiles, motivation and interests have gone through various changes, their devotion to reclusive, slow and thorough acquisition of knowledge – so characteristic of the typical figure of the classicist a century ago – has significantly diminished. On the other hand, new technologies not only became ubiquitous in the lives of students and teachers alike but also gave rise to the multidisciplinary field of Digital Humanities, and Digital Classics in particular. Several such initiatives have been developing at the Department of Classics to the University of Sofia and, as of late, have also been introduced to the BA curriculum in Classics. The paper will discuss how getting involved in activities such as encoding ancient Greek inscriptions or working on parallel text alignment can serve as a quick and efficient introduction to more complex research problems and approaches in advanced fields like epigraphy. The general request of policy-makers and public alike for practically oriented higher education yielding quick and relevant results often leads to shallower educational results of lesser quality. Solving research issues through working on ancient sources with digital tools can be a way of acquiring deeper knowledge of the subject matter of Classics more quickly. Examples from different courses and project activities will be examined in order to observe how digitally-aided research works in practice.

Explaining Media Coverage of Constitutional Court Decisions in Germany: The Role of Case Characteristics

ABSTRACT Research on the media coverage of decisions by the U.S. Supreme Court has revealed that the newsworthiness of a court decision – as determined by specific case characteristics – influences the likelihood that the decision will be covered by the media. However, in the context of European constitutional courts, media coverage of court decisions can be understood to be a form of justice reporting that focuses on general public interest and is operationalized by case characteristics that indicate the newsworthiness of the decisions. Using the case of the German Federal Constitutional Court, the determining factors behind whether court decisions are featured in the German media (i.e., that they receive media coverage in at least one newspaper article) are expounded. A local text alignment measurement of court decisions and newspaper articles enables media coverage to be identified based on original data on 3,404 decisions and 9,436 newspaper articles from between 2008 and 2018. The results demonstrate that media coverage of the German Court is not affected by declarations of unconstitutionality, though coverage is more likely when a decision is accompanied by a press release, an oral hearing, a dissenting opinion, or a combination of all three items. As these decisional aspects are the prerogative of the FCC, the results demonstrate that the German Constitutional Court has the necessary instruments to independently influence media coverage.

Cross-language text alignment: A proposed two-level matching scheme for plagiarism detection

The exponential growth of documents in various languages throughout the web, along with the availability of several editing and translation tools have made the cross-language plagiarism detection a challenging issue. Regarding its high importance, the present study focuses on the task of cross-language text alignment also known as detailed analysis which works on the outputs of the source retrieval step of cross-language plagiarism detection systems. The paper proposes a two-level matching approach with the aim of considering both syntactic and semantic information to align plagiarism fragments from the source and suspicious documents, accurately. At the first level, a vector space model which employs a multilingual word embeddings based dictionary and a local weighting technique is used in order to extract a minimal set of highly potential candidate fragment pairs rather than considering all possible pairs of fragments. This step also contains a dynamic expansion technique to cover more candidate pairs aiming at improving the system’s recall. It is followed by a more precise algorithm that examines the candidate pairs at the sentence level using a graph-of-words representation of text. As a result, by modelling both the words and their relationships, an acceptable increase in the system’s precision which is the goal of the second level is also observed. To identify evidence of plagiarism, i.e. potential cases of unauthorized text reuse, the algorithm tries to find maximum cliques from the match graph of source and suspicious texts. With this two-level investigation, the approach is capable to discriminate true plagiarism cases from the original text. The experimental results on different datasets such as PAN-PC-11, PAN-PC-12, and SemEval-2017 show that the proposed cross-language text alignment approach significantly outperforms the state-of-the-art models and can be fed into an expert system for further improvement of cross-language plagiarism detection. The source codes are publicly available on GitHub11https://github.com/MeysamRoostaee/CLPD-DetailAnalysis/archive/master.zip, for the purposes of reproducible research.

Learning Document Image Features With SqueezeNet Convolutional Neural Network

The classification of various document image classes is considered an important step towards building a modern digital library or office automation system. Convolutional Neural Network (CNN) classifiers trained with backpropagation are considered to be the current state of the art model for this task. However, there are two major drawbacks for these classifiers: the huge computational power demand for training, and their very large number of weights. Previous successful attempts at learning document image features have been based on training very large CNNs. SqueezeNet is a CNN architecture that achieves accuracies comparable to other state of the art CNNs while containing up to 50 times less weights, but never before experimented on document image classification tasks. In this research we have taken a novel approach towards learning these document image features by training on a very small CNN network such as SqueezeNet. We show that an ImageNet pretrained SqueezeNet achieves an accuracy of approximately 75 percent over 10 classes on the Tobacco-3482 dataset, which is comparable to other state of the art CNN. We then visualize saliency maps of the gradient of our trained SqueezeNet's output to input, which shows that the network is able to learn meaningful features that are useful for document classification. Previous works in this field have made no emphasis on visualizing the learned document features. The importance of features such as the existence of handwritten text, document titles, text alignment and tabular structures in the extracted saliency maps, proves that the network does not overfit to redundant representations of the rather small Tobacco-3482 dataset, which contains only 3482 document images over 10 classes.

Is your document novel? Let attention guide you. An attention-based model for document-level novelty detection

AbstractDetecting, whether a document contains sufficient new information to be deemed as novel, is of immense significance in this age of data duplication. Existing techniques for document-level novelty detection mostly perform at the lexical level and are unable to address the semantic-level redundancy. These techniques usually rely on handcrafted features extracted from the documents in a rule-based or traditional feature-based machine learning setup. Here, we present an effective approach based on neural attention mechanism to detect document-level novelty without any manual feature engineering. We contend that the simple alignment of texts between the source and target document(s) could identify the state of novelty of a target document. Our deep neural architecture elicits inference knowledge from a large-scale natural language inference dataset, which proves crucial to the novelty detection task. Our approach is effective and outperforms the standard baselines and recent work on document-level novelty detection by a margin of $\sim$ 3% in terms of accuracy.

Improving Speech to Text Alignment Based on Repetition Detection for Dysarthric Speech

Alignment of transcription to the speech finds applications in video subtitling, human–computer interaction by means of natural language communication, etc. In spite of many advancements, alignment of transcription to speech remains a challenging task and may become even more challenging for dysarthric speech. Dysarthria is a motor speech disorder resulting from damaged peripheral or central nervous system and causes slow speaking rate, pronunciation deviations, and prolonged pause interval between words and syllables. One of the problems in aligning dysarthric speech to text is the presence of repetition. Repetition can be at syllable/word/phrase level. In this work, we proposed an algorithm for syllable boundary detection followed by syllable repetition detection in dysarthric speech. When a syllable is found to be repeated, that syllable is repeated automatically in the transcription also. Modified transcription is given to the aligner along with the dysarthric speech. The proposed system when tested for word alignment with 15 utterances containing 146 words resulted in root mean square error (RMSE) of 0.138 when compared with the existing work in the literature, which gives an RMSE of 0.276.

Создание русско-тувинского параллельного подкорпуса электронного корпуса тувинского языка: первые итоги

В статье представлены предварительные итоги работ по созданию русско-тувинского параллельного подкорпуса электронного корпуса тувинского языка, которые ведутся в Научно-образовательном центре «Тюркология» Тувинского государственного университета.Рассмотрено примерное количество переводных русско-тувинских произведений, в том числе переводных произведений как зарубежных писателей, так и писателей национальных республик Советского Союза, переведенных на тувинский язык с русского языка. Авторами были использованы данные Национальной библиотеки Тувы и Тувинского государственного университета. По данным картотек библиотек насчитывается более 700 переводных произведений.В состав русско-тувинского параллельного подкорпуса включены переведенные на тувинский язык произведения М. Ю. Лермонтова «Герой нашего времени», М. А. Шолохова «Поднятая целина», В. П. Катаева «Я, сын трудового народа», А. П. Чехова «Человек в футляре» А. Фадеева «Молодая гвардия» М. Горького «Свидание», Б. Васильева «А зори здесь тихие», Е. Ильиной «Четвертая высота». На 2019 г. объем собранного русско-тувинского параллельного корпуса составляет 7631813 словоупотреблений.В работе приводятся примеры выравнивания русско-тувинских параллельных текстов, рассмотрена проблема перевода этнокультурных реалий на тувинский язык. Сравнительный анализ русско-тувинских параллельных текстов показывает степень эквивалентности и неэквивалентности перевода; трудности передачи лексики, содержащей национально-культурный компонент (слова-реалии); лексико-семантические и стилистические трансформации при переводе.

Text alignment in early printed books combining deep learning and dynamic programming

We describe a technique for transcript alignment in early printed books by using deep models in combination with dynamic programming algorithms. Two object detection models, based on Faster R-CNN, are trained to locate words. We first train an initial model to recognize generic words and hyphens by using information about the number of words in text lines. Using the model prediction on pages with a line-by-line ground-truth annotation is available, we train a second model able to detect landmark words. The alignment is then based on the identification of landmark words in pages where we only know the text corresponding to zones in the page. The proposed technique is evaluated on a publicly available digitization of the Gutenberg Bible while the transcription is based on the Vulgata, a late 4th century Latin translation of the Bible.

Japanese Pronunciation Evaluation Based on DDNN

In recent years, speech recognition technology based on deep learning model has made great progress, and the accuracy of speech recognition has reached more than 90%. In foreign language learning, speech evaluation is an important application. Billions of foreign language learners need to practice effective pronunciation. However, due to the different goals between speech recognition and speech evaluation, a single speech recognition model cannot be directly applied to pronunciation evaluation. This paper proposes a DDNN (double-layer deep neural network) model, which includes the speech text alignment model and speech recognition model. In the first layer of the speech alignment model, a new Viterbi algorithm method is proposed to find the best path for the alignment of speech and text. In the second layer of speech evaluation and scoring, we are the first to use the CNN (Convolutional Neural Network) and RNN (Recurrent Neural Network) on the encoding part of Attention. The accuracy of CTC model reaches 76.7%, and that of attention model is 81.2%. The experimental results show that the speech and text alignment method is effective, and the speech evaluation results based on the Attention model are better. The FRR (false rejection rate), FAR (false acceptance rate), and DER (diagnostic rate) in the Attention model were 4.5%, 5.1%, and 17.9%, respectively. At the same time, the evaluation of each sentence of the DDNN model in the online experiment is within 1 second, so the model can also be applied to the online real-time evaluation of speech pronunciation.

Missing Procedural Texts Repairing Based on Process Model and Activity Description Templates

Textual descriptions of business process models are used as guidance for business executives. While it makes process models much easier to understand, missing procedural text is an important practical problem. In this paper, we propose a method of repairing missing procedural text based on business process model and text alignment. The proposed method realizes the process of repairing unstructured procedural texts based on structured models. Firstly, the key concepts of alignment and repair are formally defined. Then, text repair is divided into three subtasks: alignment creation, selecting the template for activities and procedural text repairing. Each subtasks are solved in a detailed way. Experiments with Chinese recipe documents demonstrate that the proposed method is capable of repairing the missing activities of procedural text. Furthermore, the proposed method could be easily adapted to other real-world domains as it is independent on complicated manually-designed rules.

Combining Global and Local Similarity for Cross-Media Retrieval

This paper mainly studies the problem of image-text matching in order to make image and text better match. Existing cross-media retrieval methods only make use of the information of image and part of text, that is, matching the whole image with the whole sentence, or matching some image areas with some words. In order to better reveal the potential connection between image and text semantics, this paper proposes a fusion of two levels of similarity across media images-text retrieval method, constructed the cross-media two-level network to explore the better matching between images and texts, it contains two subnets for dealing with global features and local characteristics. Specifically, in this method, the image is divided into the whole picture and some image area, the text is divided into the whole sentences and words, to study respectively, to explore the full potential alignment of images and text, and then use a two-level alignment framework is used to promote each other, fusion of two kinds of similarity can learn to complete representation of cross-media retrieval. Through the experimental evaluation on Flickr30K and MS-COCO datasets, the results show that the method in this paper can make the semantic matching of image and text more accurate, and is superior to the international popular cross-media retrieval method in various evaluation indexes.

Scalable and language-independent embedding-based approach for plagiarism detection considering obfuscation type: no training phase

The efficiency and scalability of plagiarism detection systems have become a major challenge due to the vast amount of available textual data in several languages over the Internet. Plagiarism occurs in different levels of obfuscation, ranging from the exact copy of original materials to text summarization. Consequently, designed algorithms to detect plagiarism should be robust to the diverse languages and different types of obfuscation in plagiarism cases. In this paper, we employ text embedding vectors to compare similarity among documents to detect plagiarism. Word vectors are combined by a simple aggregation function to represent a text document. This representation comprises semantic and syntactic information of the text and leads to efficient text alignment among suspicious and original documents. By comparing representations of sentences in source and suspicious documents, pair sentences with the highest similarity are considered as the candidates or seeds of plagiarism cases. To filter and merge these seeds, a set of parameters, including Jaccard similarity and merging threshold, are tuned by two different approaches: offline tuning and online tuning. The offline method, which is used as the benchmark, regulates a unique set of parameters for all types of plagiarism by several trials on the training corpus. Experiments show improvements in performance by considering obfuscation type during threshold tuning. In this regard, our proposed online approach uses two statistical methods to filter outlier candidates automatically by their scale of obfuscation. By employing the online tuning approach, no distinct training dataset is required to train the system. We applied our proposed method on available datasets in English, Persian and Arabic languages on the text alignment task to evaluate the robustness of the proposed methods from the language perspective as well. As our experimental results confirm, our efficient approach can achieve considerable performance on the different datasets in various languages. Our online threshold tuning approach without any training datasets works as well as, or even in some cases better than, the training-base method.

Correction: Legal Consciousness in Action: Lay People and Accountability in the Jury Room

The original version of this article unfortunately contained mistakes in Excerpts 6 and 8 in terms of text alignment. These mistakes were not present in the proofs provided to the author and were introduced during the final stages of article production.

Attention-Passing Models for Robust and Data-Efficient End-to-End Speech Translation

Speech translation has traditionally been approached through cascaded models consisting of a speech recognizer trained on a corpus of transcribed speech, and a machine translation system trained on parallel texts. Several recent works have shown the feasibility of collapsing the cascade into a single, direct model that can be trained in an end-to-end fashion on a corpus of translated speech. However, experiments are inconclusive on whether the cascade or the direct model is stronger, and have only been conducted under the unrealistic assumption that both are trained on equal amounts of data, ignoring other available speech recognition and machine translation corpora. In this paper, we demonstrate that direct speech translation models require more data to perform well than cascaded models, and although they allow including auxiliary data through multi-task training, they are poor at exploiting such data, putting them at a severe disadvantage. As a remedy, we propose the use of end- to-end trainable models with two attention mechanisms, the first establishing source speech to source text alignments, the second modeling source to target text alignment. We show that such models naturally decompose into multi-task–trainable recognition and translation tasks and propose an attention-passing technique that alleviates error propagation issues in a previous formulation of a model with two attention stages. Our proposed model outperforms all examined baselines and is able to exploit auxiliary training data much more effectively than direct attentional models.

The Sequence Labeling Approach for Text Alignment of Plagiarism Detection

Plagiarism detection is increasingly exploiting text alignment. Text alignment involves extracting the plagiarism passages in a pair of the suspicious document and its source document. The heuristics have achieved excellent performance in text alignment. However, the further improvements of the heuristic methods mainly depends more on the experiences of experts, which makes the heuristics lack of the abilities for continuous improvements. To address this problem, machine learning maybe a proper way. Considering the position relations and the context of text segments pairs, we formalize the text alignment task as a problem of sequence labeling, improving the current methods at the model level. Especially, this paper proposes to use the probabilistic graphical model to tag the observed sequence of pairs of text segments. Hence we present the sequence labeling approach for text alignment in plagiarism detection based on Conditional Random Fields. The proposed approach is evaluated on the PAN@CLEF 2012 artificial high obfuscation plagiarism corpus and the simulated paraphrase plagiarism corpus, and compared with the methods achieved the best performance in PAN@CLEF 2012, 2013 and 2014. Experimental results demonstrate that the proposed approach significantly outperforms the state of the art methods.