Are White Ravens Ever White? - Non-Literal Adjective-Noun Phrases in Polish

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 565–573 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 565 A Corpus of Controlled Opinionated and Knowledgeable Movie Discussions for Training Neural Conversation Models Fabian Galetzka1,2, Chukwuemeka U. Eneh2, David Schlangen1 1Computational Linguistics, University of Potsdam, Germany 2Volkswagen AG fabian.galetzka@volkswagen.de Abstract Fully data driven Chatbots for non-goal oriented dialogues are known to suffer from inconsistent behaviour across their turns, stemming from a general difficulty in controlling parameters like their assumed background personality and knowledge of facts. One reason for this is the relative lack of labeled data from which personality consistency and fact usage could be learned together with dialogue behaviour. To address this, we introduce a new labeled dialogue dataset in the domain of movie discussions, where every dialogue is based on pre-specified facts and opinions. We thoroughly validate the collected dialogue for adherence of the participants to their given fact and opinion profile, and find that the general quality in this respect is high. This process also gives us an additional layer of annotation that is potentially useful for training models. We introduce as a baseline an end-to-end trained self-attention decoder model trained on this data and show that it is able to generate opinionated responses that are judged to be natural and knowledgeable and show attentiveness. Keywords: Non-Goal Driven Dialogues, Opinionated Discussions, Deep Neural Networks 1. Introduction Where dialogue modelling used to be mostly rule-based with the dialogue being driven by pre-specified knowledge representations (e.g., (Bobrow et al., 1977), (Traum and Larsson, 2003), (Stede and Schlangen, 2004)), recent years have seen efforts of basing this task on models directly learned from data. A particular strand of this research has modelled the task of producing a dialogue contribution in analogy to the translation task as one of going from one se- quence (the user utterance) to another sequence (the system utterance). The first such models solely based on data driven end-to- end approaches (Li et al., 2016; Serban et al., 2017) tended to generate universal and inconsistent utterances regarding content and personality. We illustrate this problem with the example in Figure 1, distinguishing the two consistency di- mensions knowledge (a speaker should not “forget” pre- viously known facts) and opinion (a speaker should not change their opinion, at least not without any overt trigger in the conversation). In this example, each system response is locally coherent (a good reply to its immediate precur- sor), but globally inconsistent. While this particular example is constructed, it is not very far from what these early models would have been liable to produce. One reason for this is that these models were op- timised only for local coherence, and trained from datasets such as TwitterCorpus (Ritter et al., 2010) and OpenSubti- tles corpus (Tiedemann, 2012). These datasets contain di- alogues from many people, without any information about the speakers and their opinions or knowledge state. To tackle issues like these, several augmented dialogue datasets …

Content truncated to 500 words