MCS: An In-battle Commentary System for MOBA Games

Proceedings of the 29th International Conference on Computational Linguistics, pages 2962–2967 October 12–17, 2022. 2962 MCS: An In-battle Commentary System for MOBA Games Xiaofeng Qi, Chao Li, Zhongping Liang, Jigang Liu, Cheng Zhang, Yuanxin Wei, Lin Yuan, Guang Yang, Lanxiao Huang, Min Li Tencent Timi L1 Studio, Chengdu, China sealqi,supeli,skylerliang,jigangliu,donyzhang,winsterwei, tayloryuan,mikoyang,jackiehuang,konthigali@tencent.com Abstract This paper introduces a generative system for in-battle real-time commentary in mobile MO- BA games. Event commentary is important for battles in MOBA games, which is applicable to a wide range of scenarios like live stream- ing, e-sports commentary and combat infor- mation analysis. The system takes real-time match statistics and events as input, and an ef- fective transform method is designed to con- vert match statistics and utterances into consis- tent encoding space. This paper presents the general framework and implementation detail- s of the proposed system, and provides experi- mental results on large-scale real-world match data. 1 Introduction In recent years, MOBA (Multiplayer Online Battle Arena) games have been popular all over the world. As a popular mobile MOBA game, Honor of Kings has a DAU of over 100 million, and has become an official sport in the 2022 Hangzhou Asian Games. In MOBA games, several players are divided in- to two teams and carry out a battle against each other. AI reseach in MOBA games has received many attention (Ye et al., 2020, 2022). Real-time commentary plays an important role in live stream- ing of e-sports competitions, assists the audience in the comprehension of game situation and events, expanding the presence of MOBA games. In-battle commentary aims at generating coher- ent and informative natural language descriptions based on game situation, match statistics and play- er behaviour, as illustrated in Figure 1. Similar systems have already been widely applied in real- world applications such as automated news broad- cast, television sports commentary, online video description, etc (Chen and Mooney, 2008; Wise- man et al., 2017; Nie et al., 2018; Puduppully et al., 2019; Kale and Rastogi, 2020). In this paper, we focus on the scenario of live e-sports commentary Killer hero: Shangxiang Sun Be killed hero: Yuanfang Li Event: double kill Commentary: Hero Kills Deaths Assists ⋯ Gold Gold (variation) Shangxiang Sun 12 4 7 ⋯ 14028 16.23/second Que Bian 3 2 20 ⋯ 13428 19.53/second ⋯ ⋯ ⋯ ⋯ ⋯ ⋯ ⋯ Ying Yun 8 6 8 ⋯ 15186 3.73/second Yuanfang Li 9 7 8 ⋯ 12186 3.73/second ⋯ ⋯ ⋯ ⋯ ⋯ ⋯ ⋯ 孙尚香拿下双杀，厉害！ Shangxiang Sun accomplished double kill, awesome! Figure 1: The in-battle commentary generation task. in mobile MOBA games and introduce MCS: an in-battle commentary generation system 12. Automatic commentary is grounded in in-battle match statistics and events, regarded as a data-to- text generation task. Recently, many large-scale language models have been proposed and promoted the advance in natural language generation (Rad- ford et al., 2018; Devlin et al., 2019; Radford et al., 2019). These pre-trained language models are also beneficial for data-to-text tasks (Kale and Rastogi, 2020), in this work, …

Content truncated to 500 words