Applications to Cross-genre Tasks. Due to the spike of applications that are required to maintain the conversa- tion, dialog data has recently become a popular target among researchers. The work in this field concerns problems such as learning facts through con- versation [51, 155, 62] or dialog summarization [101, 91]. More recent work in this field has focused on several inter-dialogue tasks [158, 72, 57]. Several cor- pora in the conversation domain have already been released. Cornell Movie Dialog Corpus contains a large collection of fictional conversations that have been extracted from movie scripts [40]. The Ubuntu Dialogue Corpus is a large dataset comprising 1M multi-turn dialogues with a massive number of 7M utterances [84]. The Character Mining project provides transcripts of all 9xxxxx://xxx.xxx.xxx/xxxxxx/ ten seasons of the TV show, Friends. [26]. Document retrieval has been a central task in natural language process- ing and information retrieval for a long time. The goal is to match a query against a set of documents. Previous work provided strong baselines for unstructured text retrieval and ranking problems [12, 17, 18, 6, 159, 175]. However, these systems usually assume a homogeneous domain for queries and target documents. Information extraction for the dialogue data has also been explored. Yoshino et al. presented a spoken dialogue system that ex- tracts predicate-argument structures and uses them to extract facts from news documents [170]. Flycht-Eriksson et al. developed a dialog interac- tion process of accessing textual data from a bird encyclopedia [53]. An unsupervised technique for a meeting summarization using decision-related utterances has been presented by Wang et al. [146]. Gorinski and Lapata studied movie script summarization [54]. Rosset et al. developed the Ritel project, a system designed to integrate spoken language and open-domain information retrieval [117]. All the aforementioned work uses the syntactic and semantic relation extraction in conversation domain and is relevant to this thesis. This dissertation lays the early groundwork for the document retrieval task that is focused on conversation and formal writings. Passage completion can be defined as a subtask in machine comprehension. Given a passage of text with a set of entities and a query with the masked placeholder, the goal is to select the entity from the passage that matches the query context. The CNN/Daily News dataset is a massive data set that consists of over 1M queries with their co...
