ACCESS TO MEDICAL INFORMATION THROUGH NATURAL LANGUAGE

本提案的目标是设计、实施和评估用户 基于自然语言输入的界面，将支持医疗保健 专业人员访问临床、学术和 管理目的。界面的意义在于它是 对于忙碌的专业人士来说，第一次学习很简单，很容易 断断续续的用户记忆，很一般。具体目标是： (1)为医疗信息系统的用户提供简单而通用的 用于访问信息的界面；(2)允许访问范围广泛的 信息资源；以及(3)评估用户界面 通用性、可靠性和可用性。 研究设计是每年迭代评估系统，并且 使用这些评估的结果来提高绩效。第一 第一年的部分时间将用于开发初步的自然语言 接口包括：一个工作站接口，它允许用户 以自然语言输入信息请求并审查结果；a 自然语言解析器，将分析用户的信息请求 并制作其医疗信息内容的正式表示； 和自然语言生成器，该自然语言生成器将在 简单的英语，关于系统对 请求提供信息。第一年的后半部分将用来制定一个 资源调解器，它将解释用户的请求并进行交互 以用户的名义使用信息资源。资源调解人 将提供对医学文献数据库(MEDLINE)的访问，以及 药品信息数据库(医生的案头参考)。在……里面 此外，资源调解人将向用户提供关于以下内容的反馈 对信息资源的访问限制，例如，某些数据 不可用，或需要特殊授权。 在第二年开始时，接口的可靠性将为 由MEDLINE搜索专家评估其在 解释自然语言和获取医学文献。这个 这项研究的结果将被用来增强自然语言 组件。今年剩余的时间将用于支持 用于访问临床信息数据库(CPMC)的资源中介 临床数据库)。 在接下来的三年里，可靠性研究将在 年初，使用专家提供的所有三个信息 资源。此外，每年还将进行一次评估，以 评估整个系统的可用性。这项研究将比较 卫生保健专业人员使用以下工具获取信息的能力 自然语言接口与他们获取信息的能力 使用高级访问语言(概念图)。这些研究将 用于增强自然语言组件和 资源调解人。

The goal of this proposal is to design, implement, and evaluate a user interface based on natural language input that will enable health care professionals to access information for clinical, scholarly, and administrative purposes. The significance of the interface is that it is simple for busy professionals to learn the first time, easy for intermittent users to remember, and very general. The specific aims are: (1) provide users of a medical information system a simple and general interface for accessing information; (2) enable access to a broad range of information resources; and (3) evaluate the user interface in terms of its generality, reliability, and usability. The research design is iterative - assessing the system each year, and using the results of these evaluations to improve performance. The first part of Year 1 will be used to develop the preliminary natural language interface consisting of: a workstation interface that will allow a user to enter information requests in natural language and review the results; a natural language parser that will analyze the user's information request and produce a formal representation of its medical information content; and a natural language generator that will provide feedback to the user in simple English concerning the extent of the system's understanding of the information request. The latter part of Year 1 will be used to develop a resource mediator, that will interpret the user's request and interact with information resources on the user's behalf. The resource mediator will provide access to a database of the medical literature (MEDLINE), and a database of pharmaceutical information (Physician's Desk Reference). In addition, the resource mediator will provide feedback to the user about access limitations to information resources, e.g., that certain data is not available, or requires special authorization. At the beginning of Year 2, the reliability of the interface will be assessed by having experts in MEDLINE searching rate its performance in interpreting natural language and in accessing the medical literature. The results of this study will be used to enhance the natural language components. The remainder of the year will be spent on enabling the resource mediator to access a database of clinical information (CPMC Clinical Database). In the following three years, the reliability study will be conducted in the beginning of the year, using experts for all three information resources. In addition, an evaluation will be performed each year to assess the usability of the system as a whole. This study will compare the ability of health-care professionals to access information using the natural language interface against their ability to access information using a high-level access language (conceptual graphs). These studies will be used to make enhancements in the natural language components and resource mediator.