Pun Location And Interpretation Via Deep Neural Networks

Pun is an important branch in the study of humor and is also a kind of language rhetoric.In People's Daily communication,there is a strong value of interaction.With the development of artificial intelligence,the computation of puns also gradually attracted a lot of attention of scholars in the field of natural language processing.Pun always creates a humorous effect in a certain context,in which a word implies two or more meanings by using polysemy(homographic pun)or phonological similarity to another word(heterographic pun).Therefore,the pun word plays a particularly crucial role in the computational understanding of puns.Based on the deep learning mechanism,this paper studies the location and interpretation of puns.Pun location is a task to identify the pun word in a given text,which is of great significance to understand humorous texts.Existing methods generally adopt single long sequence structure but cannot well capture the rich semantics of pun words in sentences.And,most methods of pun location are usually based on a single type of pun.Pun interpretation is a task to identify two different meanings of a pun word,which is a subsequent task of pun location.Existing methods are usually use dictionary-based,rule-based,or knowledge-based,however,these approaches do not achieve good results.In our paper,we propose two methods to solve the problem above.Our contributions can be concluded as:(1)We present an compositional semantics network with multi-task learning for pun location.This approach can considers long-distance and short-distance semantic relations between words simultaneously.For the long-distance semantic relation,we introduce multi-level embeddings to represent the most relevant aspects of the data.For the short-distance semantic relation,we exploit the complex-valued model with a self-adaptive selection mechanism based on multi-scale of input information.Meanwhile,we propose a new classification task to distinguish the homographic pun and heterographic pun.We introduce it as an auxiliary to jointly train the original pun location task,which first learns the location of different types of puns together.Experiment results show that our proposed model leads to state-of-the-art performance on both the homographic dataset and heterographic dataset.(2)We present an end-to-end neural pun interpretation model with glosses.In order to further explore the polysemy of puns,we first take the the deep learning mechanism into the interpretation of puns and combines the method of glosses.We use the bidirectional long short-term memory networks to learn the context embeddings of the pun word and the glosses embeddings of the set of senses corresponding to the pun word.Then,we calculate the similarity of contextual embeddings of the pun word and the gloss embeddings for each corresponding sense,and the final score is calculated together with the score of context embedding.We conduct experiments on the dataset of homographic pun.Compared with existing methods,experiment results show that the latest state-of-the-art results can be achieved through our model.