By Maj John T. Stamm, 445th Airlift Wing Public Affairs
/ Published June 21, 2018
Maxwell AFB, Ala. - Air University holds 3rd annual Language, Regional Expertise, and Culture (LREC) Symposium (U.S. Air Force photo by William Birchfield)
655th Intelligence, Surveillance and Reconnaissance Group Reserve Citizen Airman Tech. Sgt. Clark (first name withheld), an analyst with the 49th Intelligence Squadron, is impacting an international security analysis program.
Clark is combining deep linguistic expertise with cutting-edge data science to build a Ukrainian-English urban phrase “exonym” dictionary for the Coalition for Open-Source Defense Analysis (CODA) to use in creating an algorithm to predict the outbreak of radicalized violence in Ukraine.
Exonyms are social responses, or hostile group-level references to an opposing group, created by combining terms to create a derogatory expression. These expressions mark the use of language as an ideological weapon with the goal to mock, diminish and/or dehumanize opponents. They are being studied by two graduate students at the Naval Postgraduate School, Monterey, California, under Dr. T. Camber Warren, professor in the Department of Defense Analysis and CODA director.
“This project is an effort to use millions of geo-referenced social media messages to develop spatio-temporal maps of human discourse,” said Dr. Warren. “These maps will track the ebb and flow of local rhetoric, and will be used as inputs to statistical models to predict the location and timing of events of civil unrest and provide new approaches to situational awareness in contested environments.”
Together with the help of Dr. Marina Cobb, a Russian language expert from the Defense Language Institute Foreign Language Center, the team developed a systematic list of terms in Russian which is used to track rates of mobilization in Euro-Maidan (those who support a Ukrainian association agreement with the European Union) and anti-Maidan (those who support closer ties with Russia) groups through social-media analysis.
Clark became involved while attending the Air Force Culture and Language Center and Air University’s Language, Regional Expertise and Culture symposium at Maxwell Air Force Base, Alabama in March 2018, where Dr. Cobb moderated a presentation entitled ‘Natural Language Processing and Geospatial/Temporal Analysis.’ The annual symposium brings together hundreds of experts in the LREC community to discuss ways to further the mission.
“I discovered the researchers were missing 50 percent of their analytical goal because they did not have Ukrainian language expertise available,” Clark said. “Though American-born, I am half-Ukrainian. My earliest childhood memories include learning about my distant ancestral homeland; my first language was Ukrainian. So I gave my business card to Dr. Cobb and she put me in contact with Dr. Warren. I have assembled a team of Ukrainian-speakers at Offutt Air Force Base, Nebraska, to help refine the search terms for the algorithms to better track escalating tensions in that region.”
Dr. Warren asserts analysis of spatial aggregation of multi-lingual discourse in social media messages has been shown to enhance predictions of violent events. The key component required for such analysis is a dictionary of search terms which can be used to track ongoing conversations over space and time. If all the pieces are combined effectively, knowledge gleaned from social media can be used to gain leading indications and warnings of radicalized violence before it erupts. In regards to Ukraine, Clark’s knowledge and input is invaluable.
“Clark is helping us to expand our metrics to include Ukrainian Cyrillic exonyms used by both sides in the conflict, which we believe can greatly improve the predictive power of our statistical models,” Dr. Warren said. “This marks a substantial advance over previous efforts.”
Through a variety of similar projects, the group hopes to continue building understanding of social media trends and warning signs associated with future military conflicts.
“Our aim is to show that open-source data streams can be used to enhance situational awareness, and to provide leading indications of future violent events, in environments where alternative forms of human domain intelligence may be lacking,” said Dr. Warren. “In any case, valuable insights can be gained from analysis of patterns in which hate speech evolves.”
551 E. Maxwell Blvd, Bldg 500, Maxwell AFB, AL 36112