From the introduction: The division of text into visual segments such as sentences, paragraphs and sections achieves many functions, such as easing navigation, achieving pragmatic effect, improving readability and reflecting the organisation of information (Wright, 1983; Schriver 1997). In this paper, we report a small experiment that investigates the effect of different layout configurations on the interpretation of the antecedent of anaphoric referring expressions. Layout has so far played little role in Natural Language Generation (NLG) systems. The layout of output texts is generally very simple. At worst, it consists of only a single paragraph consisting of a few sentences; at best it is predetermined by schemas (Coch, 1996; Porter and Lester, 1997) or discourse plans (Milosavljevic, 1999). However, recent work by Power (2000) and Bouayad et al. (2000) has integrated graphically signalled segments (e.g., by whitespace, punctuation, font and face alternation) such as paragraphs, lists, text-sentences and text-clauses in a hierarchical tree-like representation called the document structure.2 This work was carried out within the ICONOCLAST project (Integrating CONstraints On Layout and Style), which aims at automatically generating formatted texts in which the formatting decisions affect the wording and vice-versa.3 If document structure affects the comprehensibility of referring expressions, this must be taken into account in any attempt to generate felicitous formatted texts. This will go a step further from current research in the automatic generation of referring expressions, where only the effect of discourse structure and grammatical function has been investigated (Dale and Reiter, 1995; Cristea et al., 1998;Walker et al., 1998; Kibble and Power, 1999).