William C. Ogden

Using Natural Language Interfaces

Publisher Summary This chapter provides an overview of natural language interfaces. A goal of human factors research with computer systems is to develop human–computer communication modes that are both error tolerant and easily learned. The goal of most natural language systems is to provide a program interface that minimizes the training required. To most, this means supplying a system that allows the use of the words and syntax of a language used in common noncomputer discourse, such as English. There is some disagreement as to the amount of understanding or flexibility that is required in the system. For example, systems have been proposed that provide natural language by permitting the user to construct English sentences by selecting words from menus. However, the idea that a system using English words in an artificial format should be considered a natural language system has been rejected.

QUILT: implementing a large-scale cross-language text retrieval system

Q~T (QWXYUaa Interfacewith Light Tmnaladon8)k a prototype implementadon of a completecmas-langusge textretrieval 8ystemtbat takeaFkgli8hqude3snd pmduceaw811glo8a translations of Spani8hdocumenla.The ayatcmindexesthe Spmiah(kmmnts inspatda but-* the B@iahquayiIltoa Spaniab equivalent aet thraugh a novel combinadoo of lexical methodsandpardkkorpus diaambwatkm. Similarmcthodaare appliedtotherctwned documontto -a-transladm thatcanbeemmined bynon-spaniah weakemtogauwhc~ vanceofthe documenttotheorkinalmquuy. lkaystem integralwtraditional,ghaary-bad machinetramhdon technologywith informationmlriev81awoacka 8nddemomtmtfxhat mhtively simpletermaubatitutionand diaambigudkn approached can be viabk for cross-languagetext refrieval.

Query languages for the casual user: Exploring the middle ground between formal and natural languages

In the past the non-programmer who wanted the information contained in a computer database had to employ an expert programmer knowledgeable in the language and structure of the database. Now languages are being developed that are designed to be used by an infrequent or “casual” user who has limited knowledge of how the data is stored or retrieved by the computer. These special purpose query languages which allow these casual users to retrieve information from computer databases are commonly referred to as “nonprocedural” (Leavenworth and Sammet, 1974) because users need only describe the data to be retrieved, not how it is to be retrieved. These languages can be classified into two basic types which are characterized by the level of constraint imposed on the syntax and vocabulary of the language (Ehrenreich, 1981). Formal query languages have a very constrained syntax and vocabulary, while natural query languages have a relatively unconstrained syntax and vocabulary. If we consider the level of constraint that can be imposed on a query language as a continuum, then formal and natural query languages represent the two ends of this continuum. There has been considerable debate over the issue of which end of this continuum best meets the needs of the casual database user (e.g. Hill, 1972; Petrick, 1976; Shneiderman, 1980). Proponents of formal languages contend that these users benefit from learning a constrained language which teaches a concise and unambiguous way of communicating with the computer. Proponents of natural languages, on the other hand, contend that more people would be able to access database information if they could use their own natural languages. Evidence from studies of the use of some of these query languages however, indicate that neither formal, nor natural languages are easy to use. These studies suggest that the casual user will have difficulty operating at either end of the level-of-constraint continuum.

IMPLICATIONS OF A COGNITIVE MODEL OF DATABASE QUERY: COMPARISON OF A NATURAL LANGUAGE, FORMAL LANGUAGE AND DIRECT MANIPULATION INTERFACE

We have developed an cognitive model of query writing, shown in Figure 1, which consists of three distinct stages. In stage one, Query Formulation, users attempt to satisfy goals by drawing on what information they may have about the stored data to generate a natural language request. In stage two, Query Translation, users must use pragmatic information about the database, including information about the systems model of the database (i.e. objects and available actions on those objects), to translate their natural language request into a request that satisfy the pragmatics of the intended query. In the third stage, Query Writing, users need information about the syntactic and semantic rules of the particular query language interface to produce a query that will be acceptable to the system.

What do users say to their natural language interface

A controlled laboratory study was conducted to evaluate a natural language database query program. Our research goal was to assess how well a commercially available natural language interface would meet the needs of a database questioner who had no formal query training. The interface was evaluated by observing seven subjects as they learned and used the product. Results show that the interface had a difficult time understanding the subjects. Only 28% of their first questions resulted in a correct result and 16% of the problems were thought to have been correctly answered by the subject but in fact had not. These results are comparable to results of evaluations of other natural language products and suggest that natural-language-interface users will need to be trained.

Text alignment in the real world: improving alignments of noisy translations using common lexical features, string matching strategies and n-gram comparisons

Alignment methods based on byte-length comparisons of alignment blocks have been remarkably successful for aligning good translations from legislative transcriptions. For noisy translations in which the parallel text of a document has significant structural differences, byte-alignment methods often do not perform well. The Pan American Health Organization (PAHO) corpus is a series of articles that were first translated by machine methods and then improved by professional translators. Many of the Spanish PAHO texts do not share formatting conventions with the corresponding English documents, refer to tables in stylistically different ways and contain extraneous information. A method based on a dynamic programming framework, but using a decision criterion derived from a combination of byte-length ratio measures, hard matching of numbers, string comparisons and n-gram co-occurrence matching substantially improves the performance of the alignment process.

A cognitive model of database querying: a tool for novice instruction

Two experiments examine the effects of incorporating user knowledge into the design of training materials for a database querying system. In Experiment 1 an informal cognitive model of a query language is derived from the verbal reports of expert users, and incorporated into existing documentation. Two groups of subjects were then asked to solve queries using either the revised or original manual. In Experiment 2 the cognitive model was formalized to explicitly describe the conceptual and procedural information that was incorporated into training materials. Three groups of subjects then received either a conceptual model, procedural model, or neither in addition to basic instructions, and then solved four sets of queries. The results show that whether or not a given type of information facilitates performance depends on the type of query, and whether the model is consistent with the operation of the query system.

Evaluating Human-Computer Dialog Styles: Command vs. Form/Fill-in for Report Modification

This study compared three human-computer interface designs for a report modification task. The designs varied as a function of dialog style. One design represented a command dialog style, another design represented a form fill-in style, and the third design was a hybrid of the two styles. The performance of computer naive subjects was recorded as they used all three interface designs to make changes to the format of a computer generated tabular report. Results indicated that subjects performed better with and preferred to use the interface based on the form fill-in style. The conclusion was that a human-computer interface designed around a form fill-in dialog style will benefit naive computer users engaged in tasks like the report formatting task, which require the modification of a small set of parameters.

Visible speech improves human language understanding: implications for speech processing systems

Evidence from the study of human language understanding is presented suggesting that our ability to perceive visible speech can greatly influence our ability to understand and remember spoken language. A view of the speakers face can greatly aid in the perception of ambiguous or noisy speech and can aid cognitive processing of speech leading to better understanding and recall. Some of these effects have been replicated using computer synthesized visual and auditory speech. Thus, it appears that when giving an interface a voice, it may be best to give it a face too.

CRL/BRANDEIS: THE DIDEROT SYSTEM

Diderot is an information extraction system built at CRL and Brandeis University over the past two years. It was produced as part of our efforts in the Tipster project. The same overall system architecture has been used for English and Japanese and for the micro-electronics and joint venture domains.