Computational models of spatial relationships: shifting the focus to cognitive plausibility

空间在人类认知中起着基础性的作用。人们经常通过自然语言、绘画和手势来交流空间。基于第一种方式的界面预计将扩大用户群体，并适应广泛使用计算机系统（例如，地理信息系统、基于内容的图像检索系统、智能轮椅）的需求。如果机器能够理解空间中物体的组织，并像人类一样在语言上交流空间，那将是不可估量的帮助。它们将为有经验的用户的日常工作提供便利；方便视障人士使用；有利于日常的非专业用户，他们不倾向于花很多时间在培训上，也不想与最先进的WIMP（窗口、图标、菜单、指针）界面作斗争。拟议的工作将是朝着这一宏伟目标迈出的又一步。在日常情况下，空间通常被视为由空间关系引起的结构，空间关系充当视觉感知数据和自然语言之间的连接纽带。这些关系的计算模型的发展已经取得了重大进展。然而，与自然语言的联系却被忽视了。简化假设会产生计算效率高的算法，但不会产生认知上合理的模型。如果一个像“北方”这样简单的单词对人和机器来说有明显不同的含义，那么两者之间的交流就会受到阻碍。自从空间分析形成以来，计算能力得到了显著提高。空间关系不仅从计算角度，而且从包括认知和语言在内的其他角度进行了彻底的研究。我们的目标是研究空间关系计算模型的发展，同时将重点转移到认知合理性上。这是朝着开发能够理解和产生对空间关系的直观的、类似人类的语言描述的系统迈出的又一重要一步。

Space plays a fundamental role in human cognition. People often communicate about space through natural language, drawings and gestures. Interfaces based on the first modality are expected to expand the community of users and suit the demand for a wide-ranging access to computer systems (e.g., geographic information systems, content-based image retrieval systems, intelligent wheelchairs). Machines that could comprehend the organization of objects in space and communicate linguistically about space, like humans do, would be immeasurably helpful. They would facilitate the routine work of the experienced user; be accessible to the visually impaired; benefit the everyday, non-expert user, who is not inclined to spend much time on training and does not want to struggle with state of the art WIMP (Windows, Icons, Menus, Pointers) interfaces. The proposed work will constitute another step towards this ambitious goal. In everyday situations, space is often viewed as a construct induced by spatial relationships---which act as a connecting link between visually perceived data and natural language. There have been significant advancements in the development of computational models of these relationships. The link with natural language, however, has been overlooked. Simplifying assumptions lead to, e.g., computationally efficient algorithms, but not to cognitively plausible models. If a word as simple as "north" has significantly different meanings for the human and the machine, communication between the two is hindered. Since the formative years of spatial analysis, computational power has dramatically improved. Spatial relationships have been thoroughly investigated not only from a computational perspective, but from other perspectives also, including cognitive and linguistic. Our goal is to investigate the development of computational models of spatial relationships while shifting the focus to cognitive plausibility. It is to take another, significant step towards the development of systems able to understand and generate intuitive, human-like linguistic descriptions of spatial relationships.