Appelt, D.E., Planning Natural Language Utterances, Proceedings of the Eighth Interns. tional Joint Conference on Artificial Intelligence, 1983, pp. 59-62.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....its ready availability. Its design is based on the assumption that division between planning the content of speech and its verbaiization is unambiguously defined. This partitioning of the generatire decisions into levels marks one aspect of difference between MUMBLE and Appelt s generation system lAppelt 83] which emphasizes the homogeneity of various deci sional procedures The input to MUMBLE consists of realization spec ifications (r specs) representing the system s communicative goal, and produced by the planner (or the strategic component) Given the input r spec, MUM BLE assigns (or, ....

Appelt, D.E., Planning Natural Language Utterances, Proceedings of the Eighth Interns. tional Joint Conference on Artificial Intelligence, 1983, pp. 59-62.

....a structure with insufficient definition of choice points, while plan operators [Hovy 88, Moore Paris 89] provided too many choice points without sufficient structure. In order to answer a question within the task at 97 hand, GENIE needed to make choice point decisions based on intentional [Appelt 85, Moore Paris 89] domain and textual [McKeown 85, Paris 87] knowledge. Previous work on text generation has tended to put the emphasis on one of the three, incorporating the other two implicitly. The problem can be best illustrated by looking at the informal description of the strategy of ....

Appelt, D. E. Planning Natural Language Utterances. Cambridge University Press, Cambridge, England, 1985.

....in mixed initiative dialogues. This presupposes the modelling of realistic conversational settings, and thus requires explicit representation of the speakers beliefs, goals and plans, as well as characterisation of appropriate dialogue behaviour by conversational roles. HAM ANS [16] and KAMP [2] were early systems in this line, PEA [14] and EDGE [4] are recent examples. Recent advances in resource bounded agency [5, 3] conflict resolution [8] and rational agency [7] have led to a third approach, focussing on conversational agents. The system models a communicating agent, albeit a ....

D. Appelt. Planning Natural Language Utterances. Cambridge University Press, Cambridge, 1985. Studies in Natural Language Processing.

.... desirability has been independently advocated in previous work: ffl It works in a two pass draft and revision mode [ Mann 1983 ] Vaughan and McDonald 1986 ] Yazdani 1987 ] Wong and Simmons 1988 ] Cline and Nutter 1991 ] Inui et al. 1992 ] ffl It interleaves planning and realization [ Appelt 1985 ] Hovy 1988 ] ffl It is incremental [ De Smedt 1990 ] ffl It uses declarative knowledge sources [ Nogier 1990 ] Polguere 1990 ] ffl It is stratificational and modular [ Polguere 1990 ] ffl It uses a uniform underlying formalism (fuf s feature structures) to represent utterances at ....

D. Appelt. Planning Natural Language Utterances. Studies in Natural Language Processing. Cambridge University Press, 1985.

....mt2.tex; 10 06 1996; 17:35; no v. p.5 6 Michael Elhadad generator will pick a combination of form and color that can be used to uniquely identify the referent and differentiate it from all other blocks in the background. Several algorithms for modifier selection are presented in [8] pp.249 262, [2], 35] 38] and [9] In particular, these algorithms ensure that selected modifiers do not trigger unwanted implicatures. In attributive usages, the goal of the speaker is to inform the hearer of some property of an object. In [31] and [2] for example, modifiers are introduced to perform inform ....

....for modifier selection are presented in [8] pp.249 262, 2] 35] 38] and [9] In particular, these algorithms ensure that selected modifiers do not trigger unwanted implicatures. In attributive usages, the goal of the speaker is to inform the hearer of some property of an object. In [31] and [2] for example, modifiers are introduced to perform inform speech acts. In this case, the generator must map from the information in the knowledge base describing the object to a modifier denoting the property being attributed. Note that in kamp [2] the notion of action subsumption was introduced ....

[Article contains additional citation context not shown here]

D. Appelt. Planning Natural Language Utterances. Studies in Natural Language Processing. Cambridge University Press, 1985.

....only one or two content units. gossip and ana (cf. table of Section 1.1.2.4) illustrate these two extremes. While both generate summaries, the former is representative of the class of microcoded generators that also includes fog [ Bourbeau et al. 1990 ] lfs [ Iordanskaja et al. 1994 ] kamp [ Appelt 1985 ] kalipsos [ Nogier 1990 ] fn [ Reiter 1991 ] epicure [ Dale 1992 ] igen [ Rubinoff 1992 ] comet [ McKeown et al. 1990 ] avdisorII [ Elhadad 1993b ] and plandoc [ Kukich et al. 1994 ] while the latter is representative of the class of macrocoded generators that also includes Danlos ....

.... analysis to acquire revision rules In order to handle supplementary content opportunistically, the new generation model described above requires the acquisition of a new type of linguistic knowledge structure: revision operations specifying the 14 Much publicized in the research literature [ Appelt 1985] [Danlos 1986 ] Meteer 1990 ] Rubinoff 1992] yet of marginal significance in most practical applications as pointed out by [ Reiter 1994 ] 10 Complex sentence S 2 containing two floating facts: Houston, TX Buck Johnson scored a season high 26 points Thursday night and the Houston ....

[Article contains additional citation context not shown here]

D. Appelt. Planning Natural Language Utterances. Studies in Natural Language Processing. Cambridge University Press, 1985.

....operators which have no effects and which can only be identified by doing the kind of reasoning which they were supposed to obviate. I will not, however, propose an alternative. This work has been extremely influential, and has led to the development of numerous very appealing systems (e.g. Appelt 1985; Grosz and Sidner 1986) The argument below is simply intended to set some limits on the use of simple minded planning algorithms in this context, and to suggest a set of research problems. Part of the motivation for the analysis presented here is that I do not believe that straightforward ....

....that the milk be poured into a jug, It seems as though we need a mediating level of description to connect the surface effect of an utterance and its intended effect in a situation. 120 Speech act theory and epistemic planning Most AI work in this area (Allen 1980, Cohen and Perrault 1979, Appelt 1985) has followed Searle (1969) and Austin (1962) in proposing a range of speech acts. A speech act is an action which is characterised partly in terms of the production of some utterance, partly in terms of changes in the knowledge and beliefs of the speaker and hearer. We might, for instance, ....

[Article contains additional citation context not shown here]

Appelt, D. (1985). Planning natural language utterances. Cambridge, UK: Cambridge University Press.

....selection, constituent ordering, and lexical selection in addition to the content delimitation and text structuring. Appelt, for example, considers what to say and how to say it to be two sides of the same problem, and he employs planning at all stages of the generation process in KAMP ( Appelt 85] Traditional AI planning problems, on the other hand, had very well defined tasks. Besides having only limited domain actions, their model of the world was often grossly simplified for the purpose of testing various planning techniques developed for their planning tasks. As a result, ....

....apparent that the task of text generation is more complex and involves more than simple application of linguistic knowledge. At the turn of the decade, research in natural language generation became very active with the introduction of new results by [McKeown 82] Cohen Perrault 79] and [Appelt 85] and since then, the scope of natural language generation task has increased drastically. Most notably, the task of determining the content of the generated text was included as a legitimate part of natural language generation task. Therefore, rather than starting from the representation of the ....

[Article contains additional citation context not shown here]

Appelt, D. E. Planning Natural Language Utterances. Cambridge University Press, Cambridge, England, 1985.

....and transforms it into the worded text according to the linguistic resources of the desired natural language. So far, several NLG systems have been constructed as parts of PhD theses such as Goldman s BABEL [12] Davey s PROTEUS [5] McDonald s MUMBLE [31] Mckeown s TEXT [33] Appelt s KAMP [1], Patten s SLANG [37] Hovy s PAULINE [16] Elhadad s SURGE [7] 1 Each of them tries to address the common problems in NLG from different perspectives. Therefore, a number of approaches to NLG have been introduced in those works. A NLG system and its approach can be characterized by: ffl the ....

....system called Penman [30] NIGLE can be characterized by having the following distinct activities: CHAPTER 2. SYSTEMIC FUNCTIONAL LINGUISTIC 17 CHOSSERS SYSTEM NETWORK REALIZATION from waiting list pick one system enter system select feature no list empty waiting take a step in chooser [1] [2] 3] execute realization st. add to waiting system list traversal complated satisfied no yes yes conditions entery Figure 2.14: A basic sequential traversal algorithm Environment: contains the representation of three kinds of knowledge ( Knowledge Base, Text Plan and Text Service) ....

D.E. Appelt. Planning Natural Language Utterances. Cambridge University Press, Cambridge, 1985.

....during the first shot, rather than being diverted by an extended dialogue. The GENIE project at Columbia has focused on how to build such an advice giving system. Instead of generating text based solely on either discourse goals [McKeown 85, Paris 87, Hovy 88, Moore Swartout 89] on intentions [Appelt 85] or on the domain [Dale 88] we found a need to combine techniques from each. We have developed an object oriented architecture in which the concepts about which we talk (domain entities) the goals that may be accomplished with them (intentions) and the rhetorical acts through which we express ....

Appelt, D. E. Planning Natural Language Utterances. Cambridge University Press, Cambridge, England, 1985.

....Introduction The planning of linguistic utterances involves knowledge of both communicative goals and linguistic rules. The division of utterance planning into two levels, what to say and how to say it, is thus natural and widely used in practical generation work, although many researchers (e.g. Appelt, 1985; Danlos, 1987; Hovy, 1988) have pointed out that the two levels interact in a way which makes it difficult to maintain a clear separation between them. Meteer (1991) refers to a generation gap between a text plan and its realisation: most planners implicitly ensure that the plan can be ....

Appelt, D. (1985). Planning Natural Language Utterances.

....thus asked to generate a single sentence of complexity comparable to opening sentences in my corpus. Finally, the proposed architecture also has a unique combination of properties whose desirability has been independently supported in previous work: ffl It interleaves planning and realization [ Appelt 1985 ] Hovy 1988 ] ffl It is incremental [ De Smedt 1990 ] ffl It is stratificational and modular [ Polguere 1990 ] ffl It uses declarative knowledge sources [ Nogier 1990 ] Polguere 1990 ] ffl It uses a uniform underlying formalism (fuf s feature structures) to represent utterances at all levels ....

D. Appelt. Planning Natural Language Utterances. Studies in Natural Language Processing. Cambridge University Press, 1985.

....by such systems are thus microcoded from words conveying only one or two content units. gossip and ana (cf. Fig. 5 p. 8) illustrate these two extremes. While both generate summaries, the former is representative of the class of microcoded generators that also includes fog [3] lfs [22] kamp [2], kalipsos [40] fn [44] epicure [5] spokesman [39] igen [51] comet [36] avdisorII [10] and plandoc [28] while the latter is representative of the class of macrocoded generators that also includes Danlos generator [6] phred [23] pauline [19] semtex [49] and weiver [21] The fact that ana ....

D. Appelt. Planning natural language utterances. Studies in natural language processing. Cambridge University Press, 1985.

....tools complement the top down regime of FUGs to control the whole content realization process. 1 Introduction Unification based formalisms (Shieber 1986) and Functional Unification Grammars (FUGs) in particular, have proved popular in text generation. In previous work (Kay 1979, McKeown 1985, Appelt 1985, Paris 1987) the input to FUGs was a fully lexicalized specification. In recent work, however, the functionality of FUGs has been extended to encompass all of content realization, including lexical choice (McKeown et al. 1990, Elhadad 1991b, Smadja 1991b, Robin 1992) In this framework, content ....

....information hiding principle between different knowledge sources. It is a way to fetch constraints lying outside the FUG on demand, only when needed by Fuf to choose between alternatives. These different points correspond to needs that have been identified in many generation systems. Telegram (Appelt 1985) implemented a mechanism where a FUG and a content planner cooperated to generate referring expressions. The external construct is a generalization of this mechanism. With Pauline, Hovy (1988) advocated interleaving pervasively content realization with content planning. With external, Fuf can ....

D. Appelt. Planning Natural Language Utterances. Studies in Natural Language Processing. Cambridge University Press, 1985.

....(e.g. 126] 111] Moore s formalism has one major advantage: since it is ultimately based on a first order meta language, it should be possible to use a first order theorem prover for it. Moore s formalism has, in fact, been implemented, by Appelt, in his natural language planning system KAMP [6], 7] However, there also seem to be a number of difficulties with the formalism. The most important are as follows: 1. The process of translating the modal language into a first order one and then theorem proving in the firstorder language is inefficient. Reichgelt observes that hard wired ....

....of the act held when the speaker performed the speech act. This tells the hearer about the cognitive state of the speaker. Such inferences are called side effects . The PBTSA enjoyed some success. For example, it was used in Appelt s celebrated natural language generation system, KAMP [6], 7] Appealing though this formulation of speech acts is, it does have its problems. One of these is that it seems to make the recognition of illocutionary force a necessary component of language use: a point stated explicitly by Appelt [7, p87] But it seems extremely unlikely that humans ....

[Article contains additional citation context not shown here]

D. E. Appelt. Planning natural language utterances. In Proceedings of the Second National Conference on Artificial Intelligence (AAAI-82), pages 59--62, Pittsburgh, PA, 1982.

....we will implement two choice factors: that of attributing salience to the various elements of the sentence, and a set of stylistic criteria for handling finegrained differences between similar words. Thus, we now briefly review the research done in NLG on these two topics. 4 See, for instance, [Appelt 1985; Novak 1988; Dale 1989] and, focusing on the notion of text cohesion and avoiding the repetition of identical noun groups, Granville 1984; Buchberger and Horacek 1988] A broader survey of discursive constraints on lexicalization, including pronominalization decisions, can be found in [Robin ....

D. Appelt. Planning natural language utterances. Cambridge, UK: Cambridge University Press, 1985.

....respect to processing strategies, including blackboard management and conflict resolution; the assignment of different subtasks to modules; the organization of the modules; and the organization of knowledge resources. This issue is discussed in [ Jakeway et al. 1996 ] In other related work [ Appelt, 1985; Meteer, 1991; Horacek, 1992 ] several sentence planning tasks are treated within the process of text planning. Rambow and Korelsky, 1992; Panaget, 1994 ] have a separate sentence planning component, but they do not separate the specific subtasks of sentence planning into distinct submodules. ....

D.E. Appelt. Planning natural language utterances. Cambridge University Press, Cambridge, England, 1985.

No context found.

Douglas E. Appelt. Planning natural-language utterances. In AAAI-82 [2], pages 59--62.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC