Representing Movie Characters in Dialogues

Proceedings of the 23rd Conference on Computational Natural Language Learning, pages 99–109 Hong Kong, China, November 3-4, 2019. c⃝2019 Association for Computational Linguistics 99 Representing Movie Characters in Dialogues Mahmoud Azab1, Noriyuki Kojima1, Jia Deng2, Rada Mihalcea1 1Computer Science and Engineering University of Michigan, Ann Arbor {mazab,kojimano,mihalcea}@umich.edu 2Department of Computer Science Princeton University jiadeng@princeton.edu Abstract We introduce a new embedding model to rep- resent movie characters and their interactions in a dialogue by encoding in the same repre- sentation the language used by these charac- ters as well as information about the other par- ticipants in the dialogue. We evaluate the per- formance of these new character embeddings on two tasks: (1) character relatedness, using a dataset we introduce consisting of a dense character interaction matrix for 4,761 unique character pairs over 22 hours of dialogue from eighteen movies; and (2) character relation classification, for fine- and coarse-grained re- lations, as well as sentiment relations. Our ex- periments show that our model significantly outperforms the traditional Word2Vec con- tinuous bag-of-words and skip-gram models, demonstrating the effectiveness of the charac- ter embeddings we introduce. We further show how these embeddings can be used in conjunc- tion with a visual question answering system to improve over previous results. 1 Introduction Understanding characters (or more broadly peo- ple) plays a critical role in the human-level in- terpretation of dialogues – be those in stories, movies, or day-to-day conversations. The verbal interaction between characters provides important information (Iyyer et al., 2016; Elson et al., 2010). In these contexts, the names of characters trigger reasoning at a much deeper level than other reg- ular words, due to the character background, be- haviors, social network, and so forth. Currently, the most commonly used word embedding models such as Word2Vec (Mikolov et al., 2013a,b) and Glove (Pennington et al., 2014) represent charac- ters using the embeddings corresponding to the to- kens used to name them. Using these models in a dialogue setting to represent the characters poses Henry: I did not know you could fly a plane. Indiana: Fly yes. Land no. Dad, you have to use the machine gun. Get it ready. Eleven o’clock! Henry: What happens at eleven o’clock? Indiana: Twelve, eleven, ten. Eleven o’clock, fire! Dad, are we hit? Henry: More or less. Son, I am sorry. They got us. Indiana: Hang on, dad. We are going in. Table 1: A snippet of conversation between two char- acters from the “Indiana Jones and the Last Crusade” movie with each dialogue turn annotated with its corre- sponding speaker name. We aim to generate embed- ding representations for “Indiana” and “Henry” in a way that captures their relation. three main issues. First, name mentions in dia- logues are sparse (Azab et al., 2018), which makes it difficult for these models to learn a good qual- ity representation for these names (Barteld, 2017). Second, in dialogues or narratives, names often do not refer to the same person, and yet these embed- dings have a single vector representation …

Content truncated to 500 words