Previous work on music generation and transformation has commonly targeted single instrument or single melody music. Here, in contrast, five music genres are used with the goal to achieve selective remixing by using domain transfer methods on spectrogram images of music. A pipeline architecture comprised of two independent generative adversarial network models was created. The first applies features from one of the genres to constant-Q transform spectrogram images to perform style transfer. The second network turns a spectrogram into a real-value tensor representation which is approximately reconstructed back into audio. The system was evaluated experimentally and through a survey. Due to the increased complexity involved in processing high sample rate music with homophonic or polyphonic audio textures, the system’s audio output was considered to be low quality, but the style transfer produced noticeable selective remixing on most of the music tracks evaluated. © 2022, The Author(s),

In contrast to most recent models that generate an entire image at once, the paper introduces a new architecture for generating images one pixel at a time using a Compositional Pattern-Producing Network (CPPN) as the generator part in a Generative Adversarial Network (GAN), allowing for effective generation of visually interesting images with artistic value, at arbitrary resolutions independent of the dimensions of the training data. The architecture, as well as accompanying (hyper-) parameters, for training CPPNs using recent GAN stabilisation techniques is shown to generalise well across many standard datasets. Rather than relying on just a latent noise vector (entangling various features with each other), mutual information maximisation is utilised to get disentangled representations, removing the requirement to use labelled data and giving the user control over the generated images. A web application for interacting with pre-trained models was also created, unique in the offered level of interactivity with an image-generating GAN.

Sentiment analysis is a circumstantial analysis of text, identifying the social sentiment to better understand the source material. The article addresses sentiment analysis of an English-Hindi and English-Bengali code-mixed textual corpus collected from social media. Code-mixing is an amalgamation of multiple languages, which previously mainly was associated with spoken language. However, social media users also deploy it to communicate in ways that tend to be somewhat casual. The coarse nature of social media text poses challenges for many language processing applications. Here, the focus is on the low predictive nature of traditional machine learners when compared to Deep Learning counterparts, including the contextual language representation model BERT (Bidirectional Encoder Representations from Transformers), on the task of extracting user sentiment from code-mixed texts. Three deep learners (a BiLSTM CNN, a Double BiLSTM and an Attention-based model) attained accuracy 20-60% greater than traditional approaches on code-mixed data, and were for comparison also tested on monolingual English data.

Work on Abusive Language Detection has tackled a wide range of subtasks and domains. As a result of this, there exists a great deal of redundancy and non-generalisability between datasets. Through experiments on cross-dataset training and testing, the paper reveals that the preconceived notion of including more non-abusive samples in a dataset (to emulate reality) may have a detrimental effect on the generalisability of a model trained on that data. Hence a hierarchical annotation model is utilised here to reveal redundancies in existing datasets and to help reduce redundancy in future efforts.

A community in social networks is generally assumed to be composed of a group of individuals with similar characteristics. Although there has been a plethora of work on understanding network topologies (edge density, clustering coefficient, etc.) within an online community, the psycho-sociological compositions of social network communities have hardly been studied. The present paper aims to analyse the communities as composition of induced psycholinguistic and sociolinguistic variables (Personalities, Values and Ethics) across individuals in social media networks. The motivation behind this analysis is to understand the behavioural characteristics at individual as well as societal level in social networks. To this end, three studies were carried out on six different datasets: three Twitter corpora, two Facebook corpora, and an Essay corpus, annotated with Values and Ethics of the users. First, experiments on creating automatic models to determine the Personality and Values of individuals by analysing their language usage and social media behaviour. Second, experiments on understanding the characteristics or blend of characteristics of individuals within an online community. Finally, generation of a map of values and ethics for India, a multi-lingual and multi-cultural country. Striking similarities to general intuitive perception could be observed, i.e., the results obtained in the study resemble our general perception about the cities/towns of India. 

To find out how users' social media behaviour and language are related to their ethical practices, the paper investigates applying Schwartz' psycholinguistic model of societal sentiment to social media text. The analysis is based on corpora collected from user essays as well as social media (Facebook and Twitter). Several experiments were carried out on the corpora to classify the ethical values of users, incorporating Linguistic Inquiry Word Count analysis, n-grams, topic models, psycholinguistic lexica, speech-acts, and nonlinguistic information, while applying a range of machine learners (Support Vector Machines, Logistic Regression, and Random Forests) to identify the best linguistic and non-linguistic features for automatic classification of values and ethics.

Active learning techniques were employed for classification of dialogue acts over two dialogue corpora, the English human-human Switchboard corpus and the Spanish human-machine Dihana corpus. It is shown clearly that active learning improves on a baseline obtained through a passive learning approach to tagging the same data sets. An error reduction of 7% was obtained on Switchboard, while a factor 5 reduction in the amount of labeled data needed for classification was achieved on Dihana. The passive Support Vector Machine learner used as baseline in itself significantly improves the state of the art in dialogue act classification on both corpora. On Switchboard it gives a 31% error reduction compared to the previously best reported result.

Multimodal conversational spoken dialogues using physical and virtual agents provide a potential interface to motivate and support users in the domain of health and fitness. The paper presents a multimodal conversational Companion system focused on health and fitness, which has both a stationary and a mobile component.

The paper describes a set of experiments involving the application of three state-of- the-art part-of-speech taggers to Ethiopian Amharic, using three different tagsets. The taggers showed worse performance than previously reported results for Eng- lish, in particular having problems with unknown words. The best results were obtained using a Maximum Entropy ap- proach, while HMM-based and SVM- based taggers got comparable results.