SFU Colloquium: Fatemah Torabi Asr

Dr. Fatemah Torabi Asr, a post-doctoral research associate in the SFU Discourse Processing Lab, will be giving a colloquium in the Linguistics Department *at Simon Fraser University.*

 

Note that this talk will be held in Room 7402 of the Robert C. Brown Hall on the SFU Burnaby Campus. See more information here.

 

Title: “News Text in the Hands of a Computational Linguist”

Abstract: News text has served as a clean, easily accessible and standard type of language data in computational linguistics research. Not only has news corpora helped computational linguists to develop automated tools for parsing, translation and many other general tasks, news text itself has always been an interesting genre of discourse to analyze, among other things, human communication behaviour. This talk will cover three research topics dealing in one way or another with news text corpora. I start with introducing my psycholinguistic PhD thesis about discourse relations where the main data was a collection of news articles from Wall Street Journals, i.e., Penn Discourse Treebank. This research was mainly focused on the general usage of discourse connectives: how authors trigger causal, contrastive and other types of relations between discourse segments and whether patterns of connective omission align with cognition and information theory. The second study is related to using large-scale news corpora for socio-political analysis of critical events such as attacks and crises. I will introduce the approach we took to define and computationally formalize the challenging concept of events based on participants (named entities) and timestamps (temporal cues) appearing in news text. Finally, I will talk about my current research at the discourse processing lab on automatic fake news detection. Our methodology to tackle missinformation in news is based on the hypothesis that deception has its own linguistic characteristics. Therefore, a fake news article should be objectively different from a legitimate news article. We are currently in the process of collecting a large dataset of fake and legitimate news articles. Using suitable discourse analysis and machine learning techniques, we have built a baseline predictive model for automatic fake news detection, which outperforms a random baseline with a significant margin. The main concern, however, is to explain the classification decisions. I will close the talk with a discussion on this issue, hoping to stimulate great ideas and get feedback for my future research!