In Japan, professional and high school baseball games are extremely popular, with live video broadcasts and textual live commentary widely available on television and the Internet. Among these, textual live commentary is particularly in high demand, as it enables users to stay updated on the status of their favorite games in real time, even when they are away from home or in situations where watching video is not feasible. In not only baseball but also other sports, such commentary is typically created and distributed by humans who watch the games live. However, keeping up with the fast pace of play and producing accurate commentary in real time is a challenging task. This paper proposes a system that, to address this issue, automatically recognizes baseball play content from video footage and generates concise summary text in real time without human intervention. The system achieved 93.1% accuracy for play type recognition, 73.17% for position recognition, and 72.09% overall accuracy. Our research aims to lay the groundwork for fully automated textual live commentary, applicable not only to professional baseball, but also to amateur and other formats of baseball games.