Extracting and Inferring Personal Attributes from Dialogue

Abstract

Personal attributes represent structured information about a person, such as their hobbies, pets, family, likes and dislikes. In this work, we introduce the tasks of extracting and inferring personal attributes from human-human dialogue. We first demonstrate the benefit of incorporating personal attributes in a social chit-chat dialogue model and task-oriented dialogue setting. Thus motivated, we propose the tasks of personal attribute extraction and inference, and then analyze the linguistic demands of these tasks. To meet these challenges, we introduce a simple and extensible model that combines an autoregressive language model utilizing constrained attribute generation with a discriminative reranker. Our model outperforms strong baselines on extracting personal attributes as well as inferring personal attributes that are not contained verbatim in utterances and instead requires commonsense reasoning and lexical inferences, which occur frequently in everyday conversation.

Extracting personal information from conversations

“…Wang et al [164] create ⟨subject, predicate, object⟩ triplets by generating predicate and object, assuming that the subject is always the speaker. The major limitation of this work is that the generation is constrained by the values present in the training data, which naturally limits the applicability of the proposed approach to slightly different conversational datasets.…”

“…The major limitation of this work is that the generation is constrained by the values present in the training data, which naturally limits the applicability of the proposed approach to slightly different conversational datasets. Wang et al [164] also make predictions on each separate sentence, making a strong assumption that each input sentence contains some personal attribute information.…”