<?xml version="1.0" standalone="yes"?>
<Paper uid="J83-3005">
  <Title>The NOMAD System: Expectation-Based Detection and Correction of Errors during Understanding of Syntactically and Semantically Ill-Formed Text 1</Title>
  <Section position="12" start_page="0" end_page="0" type="concl">
    <SectionTitle>
NOMAD OUTPUT:
</SectionTitle>
    <Paragraph position="0"> We fired bombs at an unspecified target.</Paragraph>
    <Paragraph position="1"> 4.2. Recognizing and correcting interpretation errors The four interpretation error-types given above were:  1. causal violations, 2. goal violations, 3. user confirmation failure, and 4. out-of-sequence event or object reference.  The process of detecting or correcting these error types is different in principle from the five surface types, for the simple reason that, as opposed to surface errors, which can only be attributed to the message-sender himself, there are many possible different sources of interpretation errors. In particular, some surface errors can give rise to apparent interpretation errors. To see this, recall the 'returned bombs to kashin' example above. In this case, NOMAD's default selection of a word sense for an ambiguous word ('returned') can give rise to an apparent goal violation error (delivery weapons to an enemy, as opposed to firing at an enemy). Hence, the task of blame assignment here is problematic: an early surface-processing decision of NOMAD's can give rise to an apparent later interpretation problem.</Paragraph>
    <Paragraph position="2"> Similarly, a 'user confirmation' error (that is, the user will not confirm any of the interpretations offered by NOMAD) might be due to any of a number of things: the user mistyped the original message, NOMAD made an erroneous surface-text decision, or NOMAD failed to detect a surface or interpretation problem in the text. And, a 'causal violation' error (that is, 'ship sighted overhead': ships can't fly, so the error is apparently a user error) can be due either to user errors or to NOMAD's own interpretation errors. Finally, an object or event apparently referenced out of sequence can be due to either user error or an erroneous inference by NOMAD.</Paragraph>
    <Paragraph position="3">  NOMAD has proved to be a capable analyzer of ill-formed text. Some of the standard problems of scriptand plan-based understanders have been satisfactorily addressed in NOMAD, most notably, the handling of unknown words (via the FOUL-UP mechanism); and the script-selection problem, that is, knowing which scripts to apply monitoring when they go wrong (via the mechanisms of supplanting incorrect inferences (Granger 1980), and producing a set of alternate interpretations of a text (Granger 1981a, 198282a, 1982c).</Paragraph>
    <Paragraph position="4"> The most important drawback of NOMAD is its lack of extensibility. Since the system's knowledge is mainly embedded in word-level routines, adding a new word to the system requires writing a new routine, possibly duplicating information elsewhere in NOMAD, and possibly introducing new errors into otherwiseworking NOMAD code. Any new word routine should ideally take into account interactions with all the word-level routines already present in the system; some of those routines may have to be modified in light of the new entry.</Paragraph>
    <Paragraph position="5"> In practice, we do not check every routine when a new word is added. Rather, we test the system and make corrections only when a bad interaction is found. Thus, the system is not guaranteed to be selfconsistent. Since NOMAD has more than a thousandword vocabulary, it is impractical to check the entire system when a new word is added.</Paragraph>
    <Paragraph position="6"> Encoding grammatical knowledge at the word level is also cumbersome. For example, the routine for nearly every verb makes its own checks for active or passive usage. A more centralized grammatical mechanism would eliminate this kind of redundancy. In principle, the knowledge currently encoded in the word-level routines could be made declarative (that is, stored as data), so as to be more centralized and usable by other parts of the system.</Paragraph>
    <Paragraph position="7"> 5.2. VOX: A VOcabulary eXtension system To make the NOMAD system more extensible, we are currently building a new system that uses not word-level but phrasal analysis. We call this new system VOX (for vocabulary extension system). Our goal is to make this system extensible by interaction with a user, rather than by adding to the data base programmatically. null Our ideas about phrasal analysis originate from the work on the PHRAN system (Wilensky and Arens 1982). Phrasal analysis consists of matching the input to one or more phrase-level patterns stored in a knowledge data base. When the input has been matched, it is said to be understood. Semantic actions can be associated with each phrase, so that whenever a phrase is matched to part of the input a corresponding meaning representation for the phrase may be constructed. null To extend the knowledge base of the system, we simply add new patterns to the data base. Ideally, patterns are independent entities whose interaction introduces no side effects, so that new phrases can be easily added to or removed from the data base. A working prototype of VOX is already up and running (see Granger, Meyers, Yoshii, and Taylor 1983 and Meyers 1983), incorporating syntactic and grammatical analyses, semantic analyses and blame assignment, morphological analysis, and error detection and categorization. VOX's phrase knowledge base already consists of hundreds of phrases, and is being extensively tested. Furthermore, VOX's data base can be interactively 'edited' by a trained 'tutor' to add new information, including new vocabulary, new syntactic categories and constructions, and new meanings.</Paragraph>
    <Paragraph position="8"> Hence, we hope that VOX may be a first step towards a 'trainable' language-processing system. Granger, Meyers, Yoshii, and Taylor (1983) and Meyers (1983) present extensive descriptions of the state of VOX and the theories underlying it.</Paragraph>
    <Paragraph position="9"> 5.3. Summary: Surface text and its interpretations The ability to understand text is dependent on the ability to understand what is being described in the text. Hence, a reader of English must have applicable knowledge of both the situations that may be described in texts (for example, actions, states, sequences of events, goals, methods of achieving goals, etc.), and the surface structures that appear in the language, that is, the relations between the surface order of words and phrases, and their corresponding meaning structures. The process of text understanding is the combined application of these knowledge sources as a reader proceeds through a text. This fact becomes clearest when we investigate the understanding of ill-formed texts, texts that present particular problems to a reader. The line between correct and incorrect English is often unclear, so a system that cannot handle erroneous input is of limited use.</Paragraph>
    <Paragraph position="10"> Human understanding is inherently tolerant; people are naturally able to ignore and deal with many types of errors, omissions, poor constructions, etc., and get straight to the meaning of the text. Our theories have tried to take this ability into account by including knowledge and mechanisms of error noticing and correcting as implicit parts of our process models of language understanding. NOMAD and VOX are primarily engineering applications incorporating a series of theoretical results in language understanding, including script-based and goal-based understanding, and integrated error-monitoring and supplanting during understanding. The NOMAD and VOX systems are the lat-American Journal of Computational Linguistics, Volume 9, Numbers 3-4, July-December 1983 195 Richard H. Granger The NOMAD System est in a line of 'tolerant' language understanders, beginning with FOUL-UP, all based on the use of knowledge of syntax, semantics, and pragmatics at all stages of the understanding process to cope with errors.</Paragraph>
  </Section>
class="xml-element"></Paper>