Charles River Analytics received a Small Business Innovation Research (SBIR) contract from the Department of Defense (DoD) to develop software that uses Natural Language Processing (NLP) techniques to analyze free-form text for sentiment, the Cambridge, MA-based company announced March 9. The Jargon Association for Language Analysis in Domains (JALAD) program will save the DoD time and money, as the software automates the tedious task of gathering important and relevant data from frequent surveys and other information-gathering exercises.
Both a cohesive military and mission readiness depend on personnel satisfaction; therefore, the DoD frequently polls its personnel to get an accurate picture of sentiment. “It takes a lot of work to go through these surveys manually and often on tight deadlines,” says Dr. Terry Patten, principal scientist at Charles River Analytics and principal investigator of JALAD. In addition, “the military doesn’t just want to know whether people are happy or unhappy, they want to know what specific things they are happy or unhappy about,” Patten added.
Artificial intelligence (AI) and NLP algorithms can automate the processing of free-form text, which is how personnel usually share feedback on these surveys. The problem, Patten said, is that military personnel often use jargon. To be effective, JALAD needs to detect jargon and systematically build a dictionary for future iterations.
JALAD uses statistical techniques to detect jargon, looking for words and phrases in the survey responses that are not used in general English— these might indicate military jargon. JALAD then compares the contexts in which these terms are used to those of official terms. If the context is the same, they are likely to be synonyms and get added to the dictionary.
Unit commanders need to review survey results quickly and easily, whereas researchers want to dive deeper into the data. To accommodate both sets of users, JALAD plans on including a role-tailored Human Machine Interface (HMI) to ensure that the results have the right information with the right level of detail for a given user.
Phase I of the project delivered traditional aspect sentiment analysis: Was sentiment positive or negative as it pertained to one specific thing? Among other tasks, Phase II will conduct more detailed analysis of sentiment. “For example, when people say they’re angry about something, that’s qualitatively different than saying they don’t like something, even though they’re both negative sentiments,” Patten said.
Source: Charles River Analytics
Help IC News continue to bring you breaking news from across the IC and IC contracting landscape. Join our paid subscribers today.
By 
