Metaphor processing in tweets
MetadataShow full item record
This item's downloads: 36 (view details)
Metaphor plays an important role in defining the interplay between cognition and language. Despite its fuzziness, this ubiquitous figurative device is an essential element of human communication that allows us (as humans) to better understand and, thus, communicate unfamiliar experiences and concepts in terms of familiar ones. Metaphor comprehension and understanding is a complex cognitive task that includes grasping the interaction between the underlying concepts. This is very challenging for humans, let alone computers. The last few decades have witnessed a growing interest in automating this cognitive process by introducing a wealth of ideas to model the computational recognition and comprehension of metaphors in text. Many approaches and techniques have been introduced to explore the automatic processing of different types of metaphors and the preparation of metaphor-related resources. In spite of the attention that metaphor processing has gained recently, the majority of existing approaches do not process metaphors in informal settings such as social media. Twitter offers a novel way of communication that enables users all over the world to share their thoughts and experiences. The social media content circulated on this platform through the short informal tweets poses a challenge for automatic language processing due to the unstructured nature and brevity of the text as well as the vagueness of topics. Such unique characteristics of tweets, coupled with the importance of studying metaphoric usage on social media motivated me to study metaphor processing in such a context. Metaphor processing in tweets can be beneficial in many social media analysis applications, including political discourse analysis and health communication analysis. In this thesis, I investigate the automatic processing of metaphors in tweets focusing on two main tasks, namely metaphor identification and interpretation. My aim is to improve metaphor identification to study the usage of metaphoric language in healthcare communication and political discourse in social media. Furthermore, I aim to improve metaphor interpretation to aid language learners and to enrich lexical resources. I, therefore, study various NLP and deep learning techniques to automatically identify and interpret metaphors in tweets. To the best of my knowledge, there has been no attempt to process metaphors in tweets in part due to the lack of tweet datasets annotated for linguistic metaphor. Thus, the focus of the work presented here is not only introducing models to process metaphors in tweets but also developing the necessary datasets. Overall, the work is divided into three main research themes; the first focuses on the development of metaphor annotation schemes and the preparation of datasets for both tasks. The second is concerned with the automatic identification of linguistic metaphors in tweets under a relational paradigm which explores three main ideas, namely distributional semantics, meta-embedding learning and contextual modulation. Finally, the last theme focuses on metaphor interpretation along the more complex ``definition generation'' approach, which provides full explanation of a given metaphoric expression. Experiments are conducted on the introduced datasets of tweets as well as benchmark metaphor datasets to show the effectiveness of the proposed approaches. Furthermore, the proposed datasets and the best models from this thesis will be made publicly available to facilitate research on metaphor processing in general and in tweets specifically.
This item is available under the Attribution-NonCommercial-NoDerivs 3.0 Ireland. No item may be reproduced for commercial purposes. Please refer to the publisher's URL where this is made available, or to notes contained in the item itself. Other terms may apply.
The following license files are associated with this item: