文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > A Reliable Natural Language Interface to Household Appliances
侵权投诉

A Reliable Natural Language Interface to Household Appliances

“I have always wished that my computer would be as easy to use as my telephone. My wish has come true. I no longer know how to use my telephone.” – Bjarne Stroustrop (originator of C++) As household appliances grow in complexity and sophistication, they

Can we leverage the body of knowledge accumulated in decades of studying NLIDBs to help in designing and building NLIAs? As it turns out, we are able to make a strong claim in this regard, which is our central insight: for a broad class of devices, which appears to include all household appliances, the problem of NLIA is provably reducible to the NLIDB problem. That is, given an NLIDB, a planner, and a model of an appliance, we are able to automatically generate an NLIA for the appliance. We substantiate our claim both formally and experimentally.

The remainder of the paper is organized as follows. Section 2 discusses the framework for our problem, including design requirements and significant hurdles. Section 3 presents an example that we will refer to and develop throughout the paper, and it also discusses some simplifying assumptions. In section 4, we illustrate the reduction method and state the relevant theoretical results and arguments. Section 5 describes the EXACT implementation – a working NLIA based on the reduction method – and section 6 reports on experiments measuring EXACT’s performance. Sections 7 and 8 present a discussion of directions for future research and related work in the areas of NLIDB, planning, and dialogue interfaces.

request to mute a call has to be declined. In general, when the user is not looking at the device, she misses the physical cues that suggest how the device can be used. We expect our NLIA to handle both goals and impossible requests appropriately. In the case of goals, the user need not be aware that her request translates into a sequence of appliance commands – the NLIA ought to make that transparent to the user. In the case of impossible requests, the NLIA ought to respond in a way that makes clear that the request was understood, but cannot be satisfied (e.g., as in HAL’s infamous line “I can’t do that, Dave”).

2.3 Safety

When processing a complex request with a search-based planner, one must confront the possibility that the resulting plan, while achieving a user’s goal, may have unintended and harmful consequences. Consider, for example, the goal “delete my old messages”; because our answering machine has a single command for deleting all messages (both old and new), a simple plan might have surprising and unintended consequences. Since a powerful NLIA can cause considerable havoc if not restrained, some mechanism for controlling side-effects is crucial.

In 1994, Weld and Etzioni [28] introduced the ideas of safety and tidiness in planning, ideas that were meant to restrain intelligent agents from harming people or their property. Safety constraints tell the planner never to violate certain conditions. For example, “dont-disturb(written.to.tape(f) or isa(f, file))” tells a planner that no file can be deleted unless it is already written to tape. Unfortunately, these kinds of constraints are often too strict for our domain. We do not want to tell the planner that it can never delete a message. Rather, we want to constrain the planner so that it never deletes a message unless the user tells it to.

Intuitively, one would like to tell the planner to minimize side-effects that were not requested by the user, but Weld and Etzioni recognized that obeying such a constraint is intractable in the worst case, since it requires reasoning over the (infinite) space of all plans to find the minimum. Instead, they proposed tidiness constraints, which tell the agent to restore the state of the world as much as possible to the way it was in the state before the plan started executing; “restore(compressed(file))”, for example, tells the agent’s planner to recompress as many files as possible after achieving the user’s main goal. When actions are neatly reversible, tidiness constraints often cause the planner to achieve the intuitively desirable minimization. Unfortunately, many appliances have irreversible actions, like deleting messages on an answering machine, or cooking a dish in a microwave. Thus tidiness is not well suited for our domain. Section 4.1 explains how we define and enforce a new kind of safety constraint that is a good fit for the appliance domain.

2. GUIDING PRINCIPLES

Four principles underlie our approach to designing a reliable NLIA. We insist that our NLIA be predictable, process high level goals correctly, respond appropriately to impossible requests, and generate safe plans. We explain these principles in more detail below.

2.1 Predictability

As Norman and Schneiderman have argued [17, 24], predictability is an essential feature of a user interface; without it, users will lose the essential feeling of control. Norman and Schneiderman have taken their arguments as an indictment of the intelligent user interface paradigm. However, we can take the need for predictable interfaces to heart without giving up on intelligent user interfaces. Let’s consider the issue in the context of NLIAs.

There are two main aspects to an NLIA: understanding what the user wants, and carrying out her request. If an NLIA can reliably achieve these two tasks, then the user will feel in control. As we explain below, EXACT uses a sound and complete NLIDB to understand the user’s goal, and a sound and complete planner to guide action. As section 4.2 will show, the combination of guarantees on the NLIDB and the planner yields a sound and complete NLIA. Of course, these guarantees, while helpful, are no panacea: if the request is ambiguous, some sort of clarification dialog will be necessary. However, our experimental results (section 6) provide preliminary evidence that ambiguities are rare and that EXACT is reliable and predictable in practice.

2.2 Complex And Impossible Requests

An NLIA insulates the user from the peculiarities of an appliance’s command language and obviates the unpleasant task of reading and re-reading appliance manuals. As a result, there may be some mismatch between the user’s mental model of appliance capabilities and the appliance’s exact command set. This mismatch can come in at least two flavors. First, the user’s request may require a sequence of commands to satisfy it, instead of a single command. We refer to such requests as goals. Second, the user may issue a request that is impossible to satisfy. We refer to such requests as impossible requests. For example, the KX-TC1040W phone does not have a ‘mute’ button, so a user’s

3. EXAMPLE AND SIMPLIFYING ASSUMPTIONS

In order to illustrate the ideas underlying our reduction of an NLIA to an NLIDB, we use the Panasonic KX-TC1040W telephone/answering machine as a running example throughout the paper. Our model of the phone system includes 37 actions and involves a database schema with 5 relations. The relations contain anywhere from two attributes for the answering machine messages (pictured below in table 1) to eleven attributes for the phone state. The actions include phone commands such as changing the ringer volume as well as answering machine commands and commands to access phone company services such as call forwarding.

上一页第2页

你可能喜欢

热门文档

相关文档

  • Developing a Natural Language Interface to Complex ...

    A Reliable Natural Langu... 暂无评价 8页 免费 Application of Natural L.... Developing a Natural Language Interface to Complex Data GARY G. HENDRIX, ...

  • tion of a Natural-Language Interface

    A Natural Language Inter... 暂无评价 4页 免费 A Reliable Natural Langu....Interface to Data management (EUFID) system, a n a t u r a l - l ...

  • 单片机毕设英文翻译

    Many of the seniors language has reached a level...interface circuits,to highlingt the target of ...3.In the applicationof household appliances It ...

  • 毕业设计单片机中英文资料翻译MCU Description

    A visualization of small high-level language ...interface circuits, highlight the object of ... household appliances, medical equipment, aerospace,...

  • 高级口译笔记

    reliable news 新闻/通讯稿 news release 头条新闻 ...interface 无线掌上导航器 wireless palm pilot 蜂窝...household electrical appliances 局域网 Intranet 网站...

  • 超市小件物品寄存控制系统外文翻译

    Many high-level language has reached the level ...interface circuits, highlight the object of ...In Household Appliances can be said that the ...

  • 英文翻译 2

    Many high-level language has reached the level ..., navigation systems, and household appliances. ...interface circuits, highlight the object of ...

  • AT89S52单片机毕业论文中英文5000字翻译

    code language, and as so low why is the use?..., navigation systems, and household appliances. ...interface circuits, highlight the object of ...

  • 英文文献翻译

    Many high-level language has reached the level ...interface circuits, highlight the object of ...In Household Appliances can be said that the ...

  • 单片机

    Many high-level language has reached the level ...interface circuits, highlight the object of ...In Household Appliances Can be said that the ...

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email