<?xml version="1.0" standalone="yes"?>
<Paper uid="N06-1046">
  <Title>Aggregation via Set Partitioning for Natural Language Generation</Title>
  <Section position="2" start_page="0" end_page="359" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> Aggregation is an essential component of many natural language generation systems (Reiter and Dale, 2000). The task captures a mechanism for merging together two or more linguistic structures into a single sentence. Aggregated texts tend to be more concise, coherent, and more readable overall (Dalianis, 1999; Cheng and Mellish, 2000). Compare, for example, sentence (2) in Table 1 and its nonaggregated counterpart in sentences (1a)-(1d). The difference between the fluent aggregated sentence and its abrupt and redundant alternative is striking.</Paragraph>
    <Paragraph position="1"> The benefits of aggregation go beyond making texts less stilted and repetitive. Researchers in psycholinguistics have shown that by eliminating re- null (1) a. Holocomb had an incompletion in the first quarter.</Paragraph>
    <Paragraph position="2"> b. Holocomb had another incompletion in the first quarter.</Paragraph>
    <Paragraph position="3"> c. Davis was among four San Francisco defenders.</Paragraph>
    <Paragraph position="4"> d. Holocomb threw to Davis for a leaping catch.</Paragraph>
    <Paragraph position="5"> (2) After two incompletions in the first quarter, Holcomb found Davis among four San Francisco defenders for a leaping catch.</Paragraph>
    <Paragraph position="6">  corpus of football summaries dundancy, aggregation facilitates text comprehension and recall (see Yeung (1999) and the references therein). Furthermore, Di Eugenio et al. (2005) demonstrate that aggregation can improve learning in the context of an intelligent tutoring application. In existing generation systems, aggregation typically comprises two processes: semantic grouping and sentence structuring (Wilkinson, 1995). The first process involves partitioning semantic content (usually the output of a content selection component) into disjoint sets, each corresponding to a single sentence. The second process is concerned with syntactic or lexical decisions that affect the realization of an aggregated sentence.</Paragraph>
    <Paragraph position="7"> To date, this task has involved human analysis of a domain-relevant corpus and manual development of aggregation rules (Dalianis, 1999; Shaw, 1998). The corpus analysis and knowledge engineering work in such an approach is substantial, prohibitively so in  large domains. But since corpus data is already used in building aggregation components, an appealing alternative is to try and learn the rules of semantic grouping directly from the data. Clearly, this would greatly reduce the human effort involved and ease porting generation systems to new domains.</Paragraph>
    <Paragraph position="8"> In this paper, we present an automatic method for performing the semantic grouping task. We address the following problem: given an aligned parallel corpus of sentences and their underlying semantic representations, how can we learn grouping constraints automatically? In our case the semantic content corresponds to entries from a database; however, our algorithm could be also applied to other representations such as propositions or sentence plans. We formalize semantic grouping as a set partitioning problem, where each partition corresponds to a sentence. The strength of our approach lies in its ability to capture global partitioning constraints by performing collective inference over local pair-wise assignments. This design allows us to integrate important constraints developed in symbolic approaches into an automatic aggregation framework. At a local level, pairwise constraints capture the semantic compatibility between pairs of database entries. For example, if two entries share multiple attributes, then they are likely to be aggregated. Local constraints are learned using a binary classifier that considers all pairwise combinations attested in our corpus. At a global level, we search for a semantic grouping that maximally agrees with the pairwise preferences while simultaneously satisfying constraints on the partitioning as a whole. Global constraints, for instance, could prevent the creation of overly long sentences, and, in general, control the compression rate achieved during aggregation. We encode the global inference task as an integer linear program (ILP) that can be solved using standard optimization tools.</Paragraph>
    <Paragraph position="9"> We evaluate our approach in a sports domain represented by large real-world databases containing a wealth of interrelated facts. Our aggregation algorithm model achieves an 11% F-score increase on grouping entry pairs over a greedy clustering-based model which does not utilize global information for the partitioning task. Furthermore, these results demonstrate that aggregation is amenable to an automatic treatment that does not require human involvement. null In the following section, we provide an overview of existing work on aggregation. Then, we define the learning task and introduce our approach to content grouping. Next, we present our experimental framework and data. We conclude the paper by presenting and discussing our results.</Paragraph>
  </Section>
class="xml-element"></Paper>