In this paper, we explore legal argument mining using multiple levels of
granularity. Argument mining has usually been conceptualized as a sentence
classification problem. In this work, we conceptualize argument mining as a
token-level (i.e., word-level) classification problem. We use