<?xml version="1.0" standalone="yes"?>
<Paper uid="W90-0122">
  <Title>The Computer Generation of Speech with Dlscoursally and Semantically Motivated Intonation</Title>
  <Section position="9" start_page="171" end_page="172" type="ackno">
    <SectionTitle>
Acknowledgements
</SectionTitle>
    <Paragraph position="0"> The research reported here was supported by grants from RSRE Malvem under contract no.</Paragraph>
    <Paragraph position="1"> ER1/9/4/2181/23, by the University Research Council of International Computers Ltd, and by Longman.</Paragraph>
    <Paragraph position="2"> Appendix 1 COMMUNAL is a major research project that applies and develops Systemic Functional Grammar (SFG) in a very large, fully working computer program. The acronym COMMUNAL stands for COnvivial Man-Machine Understanding through NAtural Language. The prindples underlying the project are set out in  Fawcett 1988, and an illustration of a generation is presented in Tucker 1989. A fuller (but fairly informal) overall description., incl.uding some comparison with other proiects, ts gtven ia Fawcett 1990. See also Fawcett (to appear). The project is planned to last 5 years, with around 6 researchers working on it. We finished the SUccessful Phase I in 1989, and now (May 1990) are getting under way on Phase 2 The central component of the overall system is the generator, built at Cardiff. This is called GENESYS (because it GE.NErates SYStemically).</Paragraph>
    <Paragraph position="3"> ConUdbutions from the University of Leeds in Phase 1 were to build (1) a derived probabilistiC/ parser, called the RAP (for Reafistie Annealing Parser, which develops earlier work at Leeds), and (2) the inte~reter (called REVELATION, because it reveals the 'meaning' from the 'wozding'). Each of these is a major development ia its field. But because both buiM di~ealy on the relevant aspects of GFENESYS, we can characterise the coverage of the COMMUNAL system as a whole in terms of the fxze of GENESYS.</Paragraph>
    <Paragraph position="4"> Here is a quotation and a few facts to give you a perspective on COMMUNAL at the end of Phase L McDonald, Vaughan and Pustejovsky (1987:179), in referring to the Penman project at the University of S. California, say:. 'Nigel, Penman's grammar .... is the largest systemic grammar and possibly the largest machine grammar of any kind.' Although the COMMUNAL team developed GENESYS completely independently, starting from scratch with new system networks and handIing realization in a rather different way, GENESYS already has many more systems than NigeL (This is not a criticism of Nigel; the research team have been working on other components of Penman).</Paragraph>
    <Paragraph position="5"> A major theoretical difference between the two is that the networks in GENESYS are more explicitly oriented to semantics than in Nigel. We make the assumption that the system networks in the lexicogrammar are the semantic options.</Paragraph>
    <Paragraph position="6"> GENF_.SYS has around 600 semantic systems realized in grammar (syntax and morphology, and also intonation and punctuation (see below), while Nigel has about 400 grammatical systems. But GENESYS additionally does something that the builders of Nigel would have liked to do, but from which they have so far been prevented (by the requirement of a sponsor): it integrates system networks for vocabulary with the networks realized grammaticatly. GENESYS is still growing, so that in Phase 2 we estimate that it wiU more than double the number of systems rC/~liTed in syntax and grammatical items. This should enable it to handle something approaching unrestricted syntax. COMMUNAL's first major achievement is therefore the size and scope of G~YS. The second must he seen in the wider framework of the model as a whole. It has been a long-standlng goal of ~ to build a large scale system that uses the same grammar to either generate or interlmet a sentence. (Many current systems use a different grammar for each process.) The second major achievement is to have performed this task with a very large grammar - a Systemic l~mction~d Grammar. in this case. (This will be the subject of a separate paper in the future.) A third achievement (though one less relevant in the present context) has been the development of a probabilistiC/ parsex by the Leeds part of the COMMUNAl.</Paragraph>
    <Paragraph position="7"> team.</Paragraph>
    <Paragraph position="8"> Appendix 2 'Intonation' is a term susceptible to a wide range of interpretations. It may therefore be useful to list some major aspects of the complex task of generating natural intonation that will not be discussed here. The first four are not covered because they lie outside the current goals of the COMMUNAL project, while the last two are omitted because they will be implemented (we expect) by a sister project, support for which is currently being negotiated.</Paragraph>
    <Paragraph position="9"> 1. We shall not be concerned with the high level pla,,i,PS that will tailor the text to the needs of the addressee as affected by the e.ha,,~! (e.g~ to build in greater redundancy, in the form of repetition of subject matter in planning what to express overtly, act by act). (For the general notion of tailoring, see Paris 1988.) 2. We shall not discuss variation in intonatlonal characteristics of the sort that distingahh between speakers of different dialects (geographical, social class, age, etc).</Paragraph>
    <Paragraph position="10"> 3. The same goes for individual variation, i.e. intonational idleleet.</Paragraph>
    <Paragraph position="11"> 4. We shall ignore the code of tone of voice ('angry', 'conciliatory', 'delighted', etc). At the same tlme we recognize that it is an important semiotic system in its own right, and that in the longer run the way ia which it is, as it were, superimposed on the intonation systean itsdf must be modelled. We recogniTe too the problems of drawing a firm line between tone of voice and some of the quite delicate distinctions that we shall recognise in the MOOD system (c.f.</Paragraph>
    <Paragraph position="12"> Halliday% 1970 term 'key').</Paragraph>
    <Paragraph position="13"> 5. We shaft ignore any aspect of intonational variation that does not realize meaning. For eaample, it may be that speakers introduce semantically ,,-motivated variation into the pretonic segment of an intonation unit, in order to avoid monotony (of. Hoase and Johnr.on 1987).</Paragraph>
    <Paragraph position="14"> (An alternative hypothesis, of course, might be that such variation is in fact semanticaUy motivated, but that we have not yet discovered what aspects of meaning it correlates with and how best to refer to it; this is a charaeterisrie of much interpersonal me~-;-g,) 6. We shall not be concerned here with the physical implementation of the output, but simply (if only it were ~;mpleI) with providing a written teat output marked appropriately for input to the system which will integrate it with the speech synthesis representation of the segmental phonology.</Paragraph>
  </Section>
class="xml-element"></Paper>