<?xml version="1.0" standalone="yes"?>
<Paper uid="W05-1603">
  <Title>Ten Years After: An Update on TG/2 (and Friends)</Title>
  <Section position="6" start_page="0" end_page="0" type="concl">
    <SectionTitle>
6 Conclusions and Outlook
</SectionTitle>
    <Paragraph position="0"> TG/2 has been used continuously for more than ten years.</Paragraph>
    <Paragraph position="1"> From its first appearance as a &amp;quot;template generator&amp;quot; it has evolved into a framework that accommodates both shallow template-based generation and in-depth realization tasks. In combination with the grammar development environment eGram, large grammars can be developed and maintained.</Paragraph>
    <Paragraph position="2"> They can be used by both TG/2 in Lisp and XtraGen in Java.</Paragraph>
    <Paragraph position="3"> To take up the comparison with YAG again, the most important difference is perhaps the way the rules are defined.</Paragraph>
    <Paragraph position="4"> YAG uses complex nested conditionals covering alternative verbalizations, whereas TG/2 sticks to production rules based on a context-free backbone that license local trees in a derivation. null The RAGS experiment showed that comparing TG/2 with in-depth NLG systems proves difficult. TG/2 remains shallow in that it does not support complex interrelated NLG tasks such as lexical choice, aggregation, or the generation of referring expressions.</Paragraph>
    <Paragraph position="5"> Future applications of TG/2 are geared towards presenting personalized summaries about multilingual results of question answering, generating meaningful and consistent annotations of objects in the process of modeling software, and producing user manuals for technical devices in multiple languages. null Though (or because) it has matured for more than a decade, TG/2 is alive and kicking.6</Paragraph>
  </Section>
class="xml-element"></Paper>