2021年,自然语言处理技术在超大规模预训练模型、训练范式和训练效率等方面继续实现突破,适用性和效率进一步提升,带动下游知识图谱、智能语音等技术的发展,广泛落地并赋能于金融、生物医疗、互联网应用、智慧安防等诸多场景,投融资和市场规模均迎来增长。在学术领域,研究者在进一步提升预训练语言模型性能的同时,也愈发关注模型规模加速膨胀背景下的技术可解释性、泛化性及社会性问题,以及由技术依赖产生的伦理和垄断问题。
In 2021,the breakthroughs of huge pre-training language models and training paradigms have further promoted the applying capacity and the model training efficiency of Natural Language Processing (NLP) technology,and then have led to the development of downstream technologies such as Knowledge Graph and Intelligent Speech,accelerating the applying of NLP in industries of finance,medical and healthcare,Internet,and security. As a result,both the NLP market size and the NLP investments have notably raised. Meanwhile,apart from the performance of language models,the technical accountability of ever-expanding language models,the capability of widespread industrial use,and the social concerns of AI ethnic and monopoly caused by the dependency of NLP have also been growingly concerned by researchers around the world.