Rationalization: A Neural Machine Translation Approach to Generating\n Natural Language Explanations

We introduce AI rationalization, an approach for generating explanations of\nautonomous system behavior as if a human had performed the behavior. We\ndescribe a rationalization technique that uses neural machine translation to\ntranslate internal state-action representations of an autonomous agent into\nnatural language. We evaluate our technique in the Frogger game environment,\ntraining an autonomous game playing agent to rationalize its action choices\nusing natural language. A natural language training corpus is collected from\nhuman players thinking out loud as they play the game. We motivate the use of\nrationalization as an approach to explanation generation and show the results\nof two experiments evaluating the effectiveness of rationalization. Results of\nthese evaluations show that neural machine translation is able to accurately\ngenerate rationalizations that describe agent behavior, and that\nrationalizations are more satisfying to humans than other alternative methods\nof explanation.\n