Sarcasm is widely used in social communities and e-commerce platforms, failing to detect it in natural language processing tasks leads to false positives, e.g., opinion mining and sentiment classification. Recent works have indicated that the two linguistic characteristics, sentiment and incongruity information are beneficial to sarcasm detection. However, sarcasm datasets with sentiment labels are usually unavailable, and researchers consider little semantic information while modeling incongruity. In this paper, we propose a multi-task learning framework that incorporates sentiment clues by soft sentiment labels and integrates semantic information while modeling context incongruity. Experimental results on datasets show that the model we proposed yields better performance for the sarcasm detection task with the help of sentiment clues and incongruity information.

Sarcasm detection remains a challenge for numerous Natural Language Processing (NLP) tasks, such as sentiment classification or stance prediction. Existing sarcasm detection studies attempt to capture the subtle semantic incongruity patterns by using contextual information and graph information through Graph Convolutional Networks (GCN). However, direct application of dependence may inevitably introduce noisy information and inferiorly in modeling long-distance or disconnected words in the dependency tree. To better learn the sentiment inconsistencies between terms, we propose an Affection Enhanced Relational Graph Attention network (ARGAT) by jointly considering the affective information and the dependency information. Specifically, we use Relational Graph Attention Networks (RGAT) to integrate relation information guided by a trainable matrix of relation types and synchronously use GCNs to integrate affection information explicitly donated by affective adjacency matrixes. The employment of RGAT contributes to information interaction of structural relevant word pairs with a long distance. With the enhancement of affective information, the proposed model can capture complex forms of sarcastic expressions. Experimental results on six benchmark datasets show that our proposed approach outperforms state-of-the-art sarcasm detection methods. The best-improved results of accuracy and F1 are 4.19% and 4.33%, respectively.

As massive conversation examples are easily accessible on the Internet, we are now able to organize large-scale conversation corpora to build chatbots in a data-driven manner. Multi-modal social chatbots produce conversational utterances according to both textual utterances and vision signals. Due to the difficulty of bridging different modalities, the dialogue generation model of chatbots falls into local minima that only capture the mapping between textual input and textual output, as a result, it almost ignores the non-textual signals. Further, similar to the dialogue model with plain text as input and output, the generated responses from multi-modal dialogue also lack diversity and informativeness. In this paper, to address the above issues, we propose a Multi-View Meta-Learning (MultiVML) algorithm that groups samples in multiple views and customizes generation models to different groups. We employ a multi-view clustering to group the training samples so as to attend more to the unique information in non-textual modality. Tailoring different sets of model parameters for each group boosts the genereation diversity via meta-learning. We evaluate MultiVML on two variants of the OpenViDial benchmark datasets. The experiments show that our model not only better explore the information from multiple modalities, but also excels baselines in both quality and diversity.

Automatic topic essay generation (TEG) is a controllable text generation task that aims to generate informative, diverse, and topic-consistent essays based on multiple topics. To make the generated essays of high quality, a reasonable method should consider both diversity and topic-consistency. Another essential issue is the intrinsic link of the topics, which contributes to making the essays closely surround the semantics of provided topics. However, it remains challenging for TEG to fill the semantic gap between source topic words and target output, and a more powerful model is needed to capture the semantics of given topics. To this end, we propose a pretraining-based contextual self-attention (PC-SAN) model that is built upon the seq2seq framework. For the encoder of our model, we employ a dynamic weight sum of layers from BERT to fully utilize the semantics of topics, which is of great help to fill the gap and improve the quality of the generated essays. In the decoding phase, we also transform the target-side contextual history information into the query layers to alleviate the lack of context in typical self-attention networks (SANs). Experimental results on large-scale paragraph-level Chinese corpora verify that our model is capable of generating diverse, topic-consistent text and essentially makes improvements as compare to strong baselines. Furthermore, extensive analysis validates the effectiveness of contextual embeddings from BERT and contextual history information in SANs.