Dialogue agent, a derivative of intelligent agent in the field of computational linguistics, is a computer program that is capable of generating responses and performing conversation in natural language. The field of computational linguistics is flourishing due to the intensive growth of dialogue agents; the most potential one is providing voice controlled smart personal assistant service for handsets and homes. The agents are usable, accessible but perform task-related short conversations. Non-goal-oriented dialogue agents are designed to imitate extended human–human conversations, also called as chit-chat, to provide the consumer with a satisfactory experience on the conversation quality. The design of such agents is primarily defined by a language model, unlike goal-oriented dialogue agents that employees slot based or ontology-based frameworks, hence most of the methods are data-driven. This paper surveys the current state of the art of non-goal-oriented dialogue systems specifically data-driven methods, the most prevalent being deep learning. This paper aims at (a) providing an insight of recent methods and architectures proposed for building context and modeling response along with a comprehensive review of the state of the art (b) examine the type of data set and evaluation methods available (c) present the challenges and limitation that the recent models, dataset and evaluation methods constitute.

中文翻译：

---

非目标导向对话代理：最新技术、数据集和评估

对话代理是计算语言学领域智能代理的衍生物，是一种能够生成响应并以自然语言进行对话的计算机程序。由于对话代理的密集增长，计算语言学领域正在蓬勃发展；最具潜力的是为手机和家庭提供语音控制的智能个人助理服务。代理可用、可访问但执行与任务相关的简短对话。非目标导向对话代理旨在模仿扩展的人与人对话，也称为闲聊，为消费者提供满意的对话质量体验。此类代理的设计主要由语言模型定义，与员工基于插槽或基于本体的框架的面向目标的对话代理不同，因此大多数方法都是数据驱动的。本文调查了非目标导向对话系统的最新技术，特别是数据驱动的方法，最流行的是深度学习。本文旨在 (a) 提供对最近提出的用于构建上下文和建模响应的方法和架构的见解，以及对现有技术的全面回顾 (b) 检查可用的数据集类型和评估方法 (c) 当前最近的模型、数据集和评估方法构成的挑战和限制。