Character N-gram Features Research Articles

Multi-Label Emotion Classification on Code-Mixed Text: Data and Methods

The multi-label emotion classification task aims to identify all possible emotions in a written text that best represent the author’s mental state. In recent years, multi-label emotion classification attracted the attention of researchers due to its potential applications in e-learning, health care, marketing, etc. There is a need for standard benchmark corpora to develop and evaluate multi-label emotion classification methods. The majority of benchmark corpora were developed for the English language (monolingual corpora) using tweets. However, the multi-label emotion classification problem is not explored for code-mixed text, for example, English and Roman Urdu, although the code-mixed text is widely used in Facebook posts/comments, tweets, SMS messages, particularly by the South Asian community. For filling this gap, this study presents a large benchmark corpus for the multi-label emotion classification task, which comprises 11,914 code-mixed (English and Roman Urdu) SMS messages. Each code-mixed (English and Roman Urdu) SMS message manually annotated using a set of 12 emotions, including anger, anticipation, disgust, fear, joy, love, optimism, pessimism, sadness, surprise, trust, and neutral (no emotion). As a secondary contribution, we applied and compared state-of-the-art classical machine learning (content-based methods – three word n-gram features and eight character n-gram features), deep learning (CNN, RNN, Bi-RNN, GRU, Bi-GRU, LSTM, and Bi-LSTM), and transfer learning-based methods (BERT and XLNet) on our proposed corpus. After our extensive experimentation, the best results were obtained using state-of-the-art classical machine learning methods on word uni-gram (Micro Precision = 0.67, Micro Recall = 0.54, Micro F <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sub> = 0.67) with a combination of OVR multi-label and SVC single-label machine learning algorithms. Our proposed corpus is free and publicly available for research purposes to foster research in an under-resourced language (Roman Urdu).

A Deep Learning Approach for the Romanized Tunisian Dialect Identification

Language identification is an important task in natural language processing that consists in determining the language of a given text. It has increasingly picked the interest of researchers for the past few years, especially for code-switching informal textual content. In this paper, we focus on the identification of the Romanized user-generated Tunisian dialect on the social web. We segment and annotate a corpus extracted from social media and propose a deep learning approach for the identification task. We use a Bidirectional Long Short-Term Memory neural network with Conditional Random Fields decoding (BLSTM-CRF). For word embeddings, we combine word-character BLSTM vector representation and Fast Text embeddings that takes into consideration character n-gram features. The overall accuracy obtained is 98.65%.

ProSOUL: A Framework to Identify Propaganda From Online Urdu Content

Today, the rapid dissemination of information on digital platforms has seen the emergence of information pollution such as misinformation, disinformation, fake news, and different types of propaganda. Information pollution has become a serious threat to the online digital world and has posed several challenges to social media platforms and governments around the world. In this article, we propose Propaganda Spotting in Online Urdu Language (ProSOUL) - a framework to identify content and sources of propaganda spread in the Urdu language. First, we develop a labelled dataset of 11,574 Urdu news to train the machine learning classifiers. Next, we develop the Linguistic Inquiry and Word Count (LIWC) dictionary to extract psycho-linguistic features of Urdu text. We evaluate the performance of different classifiers by varying n-gram, News Landscape (NELA), Word2Vec, and Bidirectional Encoder Representations from Transformers (BERT) features. Our results show that the combination of NELA, word n-gram, and character n-gram features outperform with 0.91 accuracy for Urdu text classification. In addition, Word2Vec embedding outperforms BERT features in classification of the Urdu text with 0.87 accuracy. Moreover, we develop and classify large scale Urdu content repositories to identify web sources spreading propaganda. Our results show that ProSOUL framework performs best for propaganda detection in the online Urdu news content compared to the general web content. To the best of our knowledge, this is the first study on the detection of propaganda content in the Urdu language.

Using character N-grams to explorediachronic change in medieval English

AbstractThis paper applies character N-grams to the study of diachronic linguistic variation in a historical language. The period selected for this initial exploratory study is medieval English, a well-studied period of great linguistic variation and language contact, whereby the efficacy of computational techniques can be examined through comparison to the wealth of thorough scholarship on medieval linguistic variation. Frequency profiles of character N-gram features were generated for several epochs in the history of English and a measure of language distance was employed to quantify the similarity between English at different stages in its history. Through this a quantification of internal change in English was achieved. Furthermore similarity between English and other medieval languages across time was measured allowing for a measurement of the well-known period of contact between English and Anglo-Norman French. This methodology is compared to traditional lexicostatistical methods and shown to be able to derive the same patterns as those derived from expert-created feature lists (i.e. Swadesh lists). The use of character N-gram profiles proved to be a flexible and useful method to study diachronic variation, allowing for the highlighting of relevant features of change. This method may be a complement to traditional qualitative examinations.

Language Independent and Multilingual Language Identification using Infinity Ngram Approach

Now days it is possible to get massive amount of multilingual digital information that are generated, propagated, exchanged, stored and accessed through the web each day across the world. Such accumulation of multilingual digital data becomes an obstacle for information acquisition. In order to tackling such difficulty language identification is the first step among many steps that are used for information acquisition. Language identification is the process of labeling given text content into corresponding language category. In past decades research works have been done in the area of language identification. However, there are issues which are not solved until: multilingual language identification, discriminating language category of very closely related languages documents and labelling the language category for very short texts like words or phrases. In this investigation, we propose an approach which able to eradicate unsolved issues of language identification (i.e. multilingual and very short texts language identification) without language barrier. In order to attain this we adopt an approach of that uses all character ngram features of given text unit (i.e. word, phrase or etc). Moreover, the proposed approach has a capability of identify the language of a text at any text unit (i.e. word, phrase, sentence or document) in both monolingual and multilingual setting. The reason behind this capability of proposed approach is due to adopting word level features, in which every words need to be classify with regard to its language category. The infinity ngram approach uses all character ngrams of text unit together in order to label the language category of given text per word level. In order to observe the effectiveness of the proposed approach four experimental techniques are conducted: pure infinity character ngram, infinity ngram with location feature and infinity ngram with sentence and document level reformulation. The experimental result indicates that an infinity ngram with location feature and along with sentence and document level reformulation achieves a promising result, which is an average F-measure of 100% at word, phrase, sentence, document level in monolingual setting. As well, for multilingual setting also attains an average F-measure of 100% for both sentence and document level, but for phrase level achieves 84.33%, 88.95% and 90.19% For Amharic, Geeze and Tigrigna respectively. Beside this, at word level achieves 83.16%, 80.96% and 85.85% for Amharic, Geeze, and Tigrigna respectively.

A classification approach for detecting cross-lingual biomedical term translations

AbstractFinding translations for technical terms is an important problem in machine translation. In particular, in highly specialized domains such as biology or medicine, it is difficult to find bilingual experts to annotate sufficient cross-lingual texts in order to train machine translation systems. Moreover, new terms are constantly being generated in the biomedical community, which makes it difficult to keep the translation dictionaries up to date for all language pairs of interest. Given a biomedical term in one language (source language), we propose a method for detecting its translations in a different language (target language). Specifically, we train a binary classifier to determine whether two biomedical terms written in two languages are translations. Training such a classifier is often complicated due to the lack of common features between the source and target languages. We propose several feature space concatenation methods to successfully overcome this problem. Moreover, we study the effectiveness of contextual and character n-gram features for detecting term translations. Experiments conducted using a standard dataset for biomedical term translation show that the proposed method outperforms several competitive baseline methods in terms of mean average precision and top-k translation accuracy.

Scripting DNA: Identifying the JavaScript programmer

The attribution of authorship is required in diverse applications, ranging from ancient novels (Shakespeare's work, Federalist papers) for historical interest to recent novels for linguistic research or even out of curiosity (Robert Galbraith alias J.K.Rowling). For this problem extensive research has resulted in effective general purpose methods. Also, for other types of text the original author needs to be discovered. Especially, we are interested in methods to identify JavaScript programmers, which can be used to reveal the offender who produced malicious software on a website. So far, for this hardly studied problem, mainly general purpose methods from natural language authorship attribution have been applied. Moreover, no suitable reference dataset is available to allow for method evaluation and method development in a supervised machine learning approach. In this work we first obtain a reference dataset of substantial size and quality. Further, we propose to extract structural features from the Abstract Syntax Tree (AST) to describe the coding style of an author. In the experiments, we show that the specifically designed features indeed improve the authorship attribution of scripting code to programmers, especially in addition to character n-gram features.

Visual Character N-Grams for Classification and Retrieval of Radiological Images

Diagnostic radiology struggles to maintain high interpretation accuracy. Retrieval of past similar cases would help the inexperienced radiologist in the interpretation process. Character n-gram model has been effective in text retrieval context in languages such as Chinese where there are no clear word boundaries. We propose the use of visual character n-gram model for representation of image for classification and retrieval purposes. Regions of interests in mammographic images are represented with the character ngram features. These features are then used as input to back-propagation neural network for classification of regions into normal and abnormal categories. Experiments on miniMIAS database show that character n-gram features are useful in classifying the regions into normal and abnormal categories. Promising classification accuracies are observed (83.33%) for fatty background tissue warranting further investigation. We argue that Classifying regions of interests would reduce the number of comparisons necessary for finding similar images from the database and hence would reduce the time required for retrieval of past similar cases.