Most harmful dialogue detection models are developed for high-resourced languages. Consequently, users who speak under-resourced languages cannot fully benefit from these models in terms of usage, development, detection and mitigation of harmful dialogue utterances. Our work aims at detecting harmful utterances in under-resourced African languages. We leverage transfer learning using pretrained models trained with multilingual embeddings to develop a cross-lingual model capable of detecting harmful content across various African languages. We first fine-tune a harmful dialogue detection model on a selected African dialogue dataset. Additionally, we fine-tune a model on a combined dataset in some African languages to develop a multilingual harmful dialogue detection model. We then evaluate the cross-lingual model’s ability to generalise to an unseen African language by performing harmful dialogue detection in an under-resourced language not present during pretraining or fine-tuning. We evaluate our models on the test datasets. We show that our best performing models achieve impressive results in terms of F1 score. Finally, we discuss the results and limitations of our work.
There has been notable progress in the development of open-domain dialogue systems (chatbots) especially with the rapid advancement of the capabilities of Large Language Models. Chatbots excel at holding conversations in a manner that keeps a user interested and engaged. However, their responses can be unsafe, as they can respond in an offensive manner or offer harmful professional advice. As a way to mitigate this issue, recent work crowdsource datasets with exemplary responses or annotate dialogue safety datasets, which are relatively scarce compared to casual dialogues. Despite the quality of data obtained from crowdsourcing, it can be expensive and time consuming. This work proposes an effective pipeline, using information retrieval, to automatically repurpose existing dialogue datasets for safe chatbot development, as a way to address the aforementioned challenges. We select an existing dialogue dataset, revise its unsafe responses, as a way to obtain a dataset with safer responses to unsafe user inputs. We then fine-tune dialogue models on the original and revised datasets and generate responses to evaluate the safeness of the models.
Despite advancements in neural machine translation, word sense disambiguation remains challenging, particularly with limited textual context. Multimodal Machine Translation enhances text-only models by integrating visual information, but its impact varies across translations. This study focuses on ambiguous sentences to investigate the effectiveness of utilizing visual information. By prioritizing these sentences, which benefit from visual cues, we aim to enhance hybrid multimodal and text-only translation approaches. We utilize Latent Semantic Analysis and Sentence-BERT to extract context vectors from the British National Corpus, enabling the assessment of semantic diversity. Our approach enhances translation quality for English-German and English-French on Multi30k, assessed through metrics including BLEU, chrF2, and TER.
Causal reasoning is a critical component of human cognition and is required across a range of question-answering (QA) tasks (such as abductive reasoning, commonsense QA, and procedural reasoning). Research on causal QA has been underdefined, task-specific, and limited in complexity. Recent advances in foundation language models (such as BERT, ERNIE, and T5) have shown the efficacy of pre-trained models across diverse QA tasks. However, there is limited research exploring the causal reasoning capabilities of those language models and no standard evaluation benchmark. To unify causal QA research, we propose CALM-Bench, a multi-task benchmark for evaluating causality-aware language models (CALM). We present a standardized definition of causal QA tasks and show empirically that causal reasoning can be generalized and transferred across different QA tasks. Additionally, we share a strong multi-task baseline model which outperforms single-task fine-tuned models on the CALM-Bench tasks.
Recent studies in Multimodal Machine Translation (MMT) have explored the use of visual information in a multimodal setting to analyze its redundancy with textual information. The aim of this work is to develop a more effective approach to incorporating relevant visual information into the translation process and improve the overall performance of MMT models. This paper proposes an object-level filtering approach in Multimodal Machine Translation, where the approach is applied to object regions extracted from an image to filter out irrelevant objects based on the image captions to be translated. Using the filtered image helps the model to consider only relevant objects and their relative locations to each other. Different matching methods, including string matching and word embeddings, are employed to identify relevant objects. Gaussian blurring is used to soften irrelevant objects from the image and to evaluate the effect of object filtering on translation quality. The performance of the filtering approaches was evaluated on the Multi30K dataset in English to German, French, and Czech translations, based on BLEU, ChrF2, and TER metrics.
Knowledge-grounded dialogue systems utilise external knowledge such as knowledge graphs to generate informative and appropriate responses. A crucial challenge of such systems is to select facts from a knowledge graph pertinent to the dialogue context for response generation. This fact selection can be formulated as path traversal over a knowledge graph conditioned on the dialogue context. Such paths can originate from facts mentioned in the dialogue history and terminate at the facts to be mentioned in the response. These walks, in turn, provide an explanation of the flow of the conversation. This work proposes KG-CRuSE, a simple, yet effective LSTM based decoder that utilises the semantic information in the dialogue history and the knowledge graph elements to generate such paths for effective conversation explanation. Extensive evaluations showed that our model outperforms the state-of-the-art models on the OpenDialKG dataset on multiple metrics.
Multimodal Neural Machine Translation is focusing on using visual information to translate sentences in the source language into the target language. The main idea is to utilise information from visual modalities to promote the output quality of the text-based translation model. Although the recent multimodal strategies extract the most relevant visual information in images, the effectiveness of using visual information on translation quality changes based on the text dataset. Due to this, this work studies the impact of leveraging visual information in multimodal translation models of ambiguous sentences. Our experiments analyse the Multi30k evaluation dataset and calculate ambiguity scores of sentences based on the WordNet hierarchical structure. To calculate the ambiguity of a sentence, we extract the ambiguity scores for all nouns based on the number of senses in WordNet. The main goal is to find in which sentences, visual content can improve the text-based translation model. We report the correlation between the ambiguity scores and translation quality extracted for all sentences in the English-German dataset.
We describe initial work in developing a methodology for the automatic generation of a conversational agent or ‘chatbot’ through term and relation extraction from a relevant corpus of language data. We develop our approach in the domain of industrial heritage in the 18th and 19th centuries, and more specifically on the industrial history of canals and mills in Ireland. We collected a corpus of relevant newspaper reports and Wikipedia articles, which we deemed representative of a layman’s understanding of this topic. We used the Saffron toolkit to extract relevant terms and relations between the terms from the corpus and leveraged the extracted knowledge to query the British Library Digital Collection and the Project Gutenberg library. We leveraged the extracted terms and relations in identifying possible answers for a constructed set of questions based on the extracted terms, by matching them with sentences in the British Library Digital Collection and the Project Gutenberg library. In a final step, we then took this data set of question-answer pairs to train a chatbot. We evaluate our approach by manually assessing the appropriateness of the generated answers for a random sample, each of which is judged by four annotators.
The task of causal question answering aims to reason about causes and effects over a provided real or hypothetical premise. Recent approaches have converged on using transformer-based language models to solve question answering tasks. However, pretrained language models often struggle when external knowledge is not present in the premise or when additional context is required to answer the question. To the best of our knowledge, no prior work has explored the efficacy of augmenting pretrained language models with external causal knowledge for multiple-choice causal question answering. In this paper, we present novel strategies for the representation of causal knowledge. Our empirical results demonstrate the efficacy of augmenting pretrained models with external causal knowledge. We show improved performance on the COPA (Choice of Plausible Alternatives) and WIQA (What If Reasoning Over Procedural Text) benchmark tasks. On the WIQA benchmark, our approach is competitive with the state-of-the-art and exceeds it within the evaluation subcategories of In-Paragraph and Out-of-Paragraph perturbations.
This paper describes the submission by NUIG-DSI to the GEM benchmark 2021. We participate in the modeling shared task where we submit outputs on four datasets for data-to-text generation, namely, DART, WebNLG (en), E2E and CommonGen. We follow an approach similar to the one described in the GEM benchmark paper where we use the pre-trained T5-base model for our submission. We train this model on additional monolingual data where we experiment with different masking strategies specifically focused on masking entities, predicates and concepts as well as a random masking strategy for pre-training. In our results we find that random masking performs the best in terms of automatic evaluation metrics, though the results are not statistically significantly different compared to other masking strategies.
A meme is a form of media that spreads an idea or emotion across the internet. As posting meme has become a new form of communication of the web, due to the multimodal nature of memes, postings of hateful memes or related events like trolling, cyberbullying are increasing day by day. Hate speech, offensive content and aggression content detection have been extensively explored in a single modality such as text or image. However, combining two modalities to detect offensive content is still a developing area. Memes make it even more challenging since they express humour and sarcasm in an implicit way, because of which the meme may not be offensive if we only consider the text or the image. Therefore, it is necessary to combine both modalities to identify whether a given meme is offensive or not. Since there was no publicly available dataset for multimodal offensive meme content detection, we leveraged the memes related to the 2016 U.S. presidential election and created the MultiOFF multimodal meme dataset for offensive content detection dataset. We subsequently developed a classifier for this task using the MultiOFF dataset. We use an early fusion technique to combine the image and text modality and compare it with a text- and an image-only baseline to investigate its effectiveness. Our results show improvements in terms of Precision, Recall, and F-Score. The code and dataset for this paper is published in https://github.com/bharathichezhiyan/Multimodal-Meme-Classification-Identifying-Offensive-Content-in-Image-and-Text
In this paper, we present the NUIG system at the TIAD shard task. This system includes graph-based metrics calculated using novel algorithms, with an unsupervised document embedding tool called ONETA and an unsupervised multi-way neural machine translation method. The results are an improvement over our previous system and produce the highest precision among all systems in the task as well as very competitive F-Measure results. Incorporating features from other systems should be easy in the framework we describe in this paper, suggesting this could very easily be extended to an even stronger result.
Social media are interactive platforms that facilitate the creation or sharing of information, ideas or other forms of expression among people. This exchange is not free from offensive, trolling or malicious contents targeting users or communities. One way of trolling is by making memes, which in most cases combines an image with a concept or catchphrase. The challenge of dealing with memes is that they are region-specific and their meaning is often obscured in humour or sarcasm. To facilitate the computational modelling of trolling in the memes for Indian languages, we created a meme dataset for Tamil (TamilMemes). We annotated and released the dataset containing suspected trolls and not-troll memes. In this paper, we use the a image classification to address the difficulties involved in the classification of troll memes with the existing methods. We found that the identification of a troll meme with such an image classifier is not feasible which has been corroborated with precision, recall and F1-score.
Conversational recommender systems focus on the task of suggesting products to users based on the conversation flow. Recently, the use of external knowledge in the form of knowledge graphs has shown to improve the performance in recommendation and dialogue systems. Information from knowledge graphs aids in enriching those systems by providing additional information such as closely related products and textual descriptions of the items. However, knowledge graphs are incomplete since they do not contain all factual information present on the web. Furthermore, when working on a specific domain, knowledge graphs in its entirety contribute towards extraneous information and noise. In this work, we study several subgraph construction methods and compare their performance across the recommendation task. We incorporate pre-trained embeddings from the subgraphs along with positional embeddings in our models. Extensive experiments show that our method has a relative improvement of at least 5.62% compared to the state-of-the-art on multiple metrics on the recommendation task.
Data-to-text generation has recently seen a move away from modular and pipeline architectures towards end-to-end architectures based on neural networks. In this work, we employ knowledge graph embeddings and explore their utility for end-to-end approaches in a data-to-text generation task. Our experiments show that using knowledge graph embeddings can yield an improvement of up to 2 – 3 BLEU points for seen categories on the WebNLG corpus without modifying the underlying neural network architecture.
This paper describes the system submitted by NUIG-DSI to the WebNLG+ challenge 2020 in the RDF-to-text generation task for the English language. For this challenge, we leverage transfer learning by adopting the T5 model architecture for our submission and fine-tune the model on the WebNLG+ corpus. Our submission ranks among the top five systems for most of the automatic evaluation metrics achieving a BLEU score of 51.74 over all categories with scores of 58.23 and 45.57 across seen and unseen categories respectively.
Bilingual lexicons are a vital tool for under-resourced languages and recent state-of-the-art approaches to this leverage pretrained monolingual word embeddings using supervised or semi-supervised approaches. However, these approaches require cross-lingual information such as seed dictionaries to train the model and find a linear transformation between the word embedding spaces. Especially in the case of low-resourced languages, seed dictionaries are not readily available, and as such, these methods produce extremely weak results on these languages. In this work, we focus on the Dravidian languages, namely Tamil, Telugu, Kannada, and Malayalam, which are even more challenging as they are written in unique scripts. To take advantage of orthographic information and cognates in these languages, we bring the related languages into a single script. Previous approaches have used linguistically sub-optimal measures such as the Levenshtein edit distance to detect cognates, whereby we demonstrate that the longest common sub-sequence is linguistically more sound and improves the performance of bilingual lexicon induction. We show that our approach can increase the accuracy of bilingual lexicon induction methods on these languages many times, making bilingual lexicon induction approaches feasible for such under-resourced languages.
This work addresses the classification problem defined by sub-task A (English only) of the OffensEval 2020 challenge. We used a semi-supervised approach to classify given tweets into an offensive (OFF) or not-offensive (NOT) class. As the OffensEval 2020 dataset is loosely labelled with confidence scores given by unsupervised models, we used last year’s offensive language identification dataset (OLID) to label the OffensEval 2020 dataset. Our approach uses a pseudo-labelling method to annotate the current dataset. We trained four text classifiers on the OLID dataset and the classifier with the highest macro-averaged F1-score has been used to pseudo label the OffensEval 2020 dataset. The same model which performed best amongst four text classifiers on OLID dataset has been trained on the combined dataset of OLID and pseudo labelled OffensEval 2020. We evaluated the classifiers with precision, recall and macro-averaged F1-score as the primary evaluation metric on the OLID and OffensEval 2020 datasets. This work is licensed under a Creative Commons Attribution 4.0 International Licence. Licence details: http://creativecommons.org/licenses/by/4.0/.
Depression and anxiety are the two most prevalent mental health disorders worldwide, impacting the lives of millions of people each year. In this work, we develop and evaluate a multilabel, multidimensional deep neural network designed to predict PHQ-4 scores based on individuals written text. Our system outperforms random baseline metrics and provides a novel approach to how we can predict psychometric scores from written text. Additionally, we explore how this architecture can be applied to analyse social media data.
Wordnets are extensively used in natural language processing, but the current approaches for manually building a wordnet from scratch involves large research groups for a long period of time, which are typically not available for under-resourced languages. Even if wordnet-like resources are available for under-resourced languages, they are often not easily accessible, which can alter the results of applications using these resources. Our proposed method presents an expand approach for improving and generating wordnets with the help of machine translation. We apply our methods to improve and extend wordnets for the Dravidian languages, i.e., Tamil, Telugu, Kannada, which are severly under-resourced languages. We report evaluation results of the generated wordnet senses in term of precision for these languages. In addition to that, we carried out a manual evaluation of the translations for the Tamil language, where we demonstrate that our approach can aid in improving wordnet resources for under-resourced Dravidian languages.
Machine translation between closely related languages is less challenging and exibits a smaller number of translation errors than translation between distant languages, but there are still obstacles which should be addressed in order to improve such systems. This work explores the obstacles for machine translation systems between closely related South Slavic languages, namely Croatian, Serbian and Slovenian. Statistical systems for all language pairs and translation directions are trained using parallel texts from different domains, however mainly on spoken language i.e. subtitles. For translation between Serbian and Croatian, a rule-based system is also explored. It is shown that for all language pairs and translation systems, the main obstacles are differences between structural properties.
In recent years, several end-to-end online translation systems have been proposed to successfully incorporate human post-editing feedback in the translation workflow. The performance of these systems in a multi-domain translation environment (involving different text genres, post-editing styles, machine translation systems) within the automatic post-editing (APE) task has not been thoroughly investigated yet. In this work, we show that when used in the APE framework the existing online systems are not robust towards domain changes in the incoming data stream. In particular, these systems lack in the capability to learn and use domain-specific post-editing rules from a pool of multi-domain data sets. To cope with this problem, we propose an online learning framework that generates more reliable translations with significantly better quality as compared with the existing online and batch systems. Our framework includes: i) an instance selection technique based on information retrieval that helps to build domain-specific APE systems, and ii) an optimization procedure to tune the feature weights of the log-linear model that allows the decoder to improve the post-editing quality.
We present a freely available corpus containing source language texts from different domains along with their automatically generated translations into several distinct morphologically rich languages, their post-edited versions, and error annotations of the performed post-edit operations. We believe that the corpus will be useful for many different applications. The main advantage of the approach used for creation of the corpus is the fusion of post-editing and error classification tasks, which have usually been seen as two independent tasks, although naturally they are not. We also show benefits of coupling automatic and manual error classification which facilitates the complex manual error annotation task as well as the development of automatic error classification tools. In addition, the approach facilitates annotation of language pair related issues.
We describe IRIS, a statistical machine translation (SMT) system for translating from English into Irish and vice versa. Since Irish is considered an under-resourced language with a limited amount of machine-readable text, building a machine translation system that produces reasonable translations is rather challenging. As translation is a difficult task, current research in SMT focuses on obtaining statistics either from a large amount of parallel, monolingual or other multilingual resources. Nevertheless, we collected available English-Irish data and developed an SMT system aimed at supporting human translators and enabling cross-lingual language technology tasks.
Princeton WordNet is one of the most important resources for natural language processing, but is only available for English. While it has been translated using the expand approach to many other languages, this is an expensive manual process. Therefore it would be beneficial to have a high-quality automatic translation approach that would support NLP techniques, which rely on WordNet in new languages. The translation of wordnets is fundamentally complex because of the need to translate all senses of a word including low frequency senses, which is very challenging for current machine translation approaches. For this reason we leverage existing translations of WordNet in other languages to identify contextual information for wordnet senses from a large set of generic parallel corpora. We evaluate our approach using 10 translated wordnets for European languages. Our experiment shows a significant improvement over translation without any contextual information. Furthermore, we evaluate how the choice of pivot languages affects performance of multilingual word sense disambiguation.
In this paper, we address the problem of extracting and integrating bilingual terminology into a Statistical Machine Translation (SMT) system for a Computer Aided Translation (CAT) tool scenario. We develop a framework that, taking as input a small amount of parallel in-domain data, gathers domain-specific bilingual terms and injects them in an SMT system to enhance the translation productivity. Therefore, we investigate several strategies to extract and align bilingual terminology, and to embed it into the SMT. We compare two embedding methods that can be easily used at run-time without altering the normal activity of an SMT system: XML markup and the cache-based model. We tested our framework on two different domains showing improvements up to 15% BLEU score points.