[Paper Review] Presenting Punctuation

Michael White|ArXiv.org|Jun 10, 1995

Discourse Analysis in Language Studies9 references21 citations

TL;DR

This paper proposes a stratificational approach to punctuation realization in natural language generation (NLG), extending Nunberg's (1990) theory of punctuation absorption and transposition within the Meaning-Text framework. It introduces a structured account that models punctuation and formatting as syntactic and morphological features, enabling robust, rule-ordered realization through feature propagation and graphic rendering, with key results showing consistent handling of complex cases like embedded quotes, dashes, and comma absorption.

ABSTRACT

Until recently, punctuation has received very little attention in the linguistics and computational linguistics literature. Since the publication of Nunberg's (1990) monograph on the topic, however, punctuation has seen its stock begin to rise: spurred in part by Nunberg's ground-breaking work, a number of valuable inquiries have been subsequently undertaken, including Hovy and Arens (1991), Dale (1991), Pascual (1993), Jones (1994), and Briscoe (1994). Continuing this line of research, I investigate in this paper how Nunberg's approach to presenting punctuation (and other formatting devices) might be incorporated into NLG systems. Insofar as the present paper focuses on the proper syntactic treatment of punctuation, it differs from these other subsequent works in that it is the first to examine this issue from the generation perspective.

Motivation & Objective

• To address the lack of syntactic treatment of punctuation in NLG systems, particularly the absence of formal mechanisms for handling punctuation's structural role.
• To evaluate and improve upon Nunberg’s (1990) account of punctuation absorption and transposition from an NLG generation perspective.
• To develop a stratificational model that integrates punctuation and formatting into the syntactic and morphological realization pipeline.
• To demonstrate how punctuation phenomena—such as comma absorption, bracket absorption, and quote transposition—can be systematically realized using feature structures and rule-based propagation.
• To instantiate the model within the CoGenTex core realizer using the Meaning-Text framework, ensuring compatibility with existing NLG architectures.

Proposed method

• Uses a stratificational architecture with five levels: Discourse Syntactic Representation (DSyntR), Surface Syntactic Representation (SSyntR), Discourse Morphological Representation (DMorphR), Surface Morphological Representation (SMorphR), and Discourse Graphical Representation (DGraphR).
• Models punctuation and formatting as features (e.g., [btw:comma], [btw:dash], [font:ital]) attached to syntactic nodes, with phrasal and lexical attributes distinguished.
• Applies rule-based propagation: e.g., a rule inserts commas around descriptive attributes unless overridden by stronger punctuation, such as a dash.
• Handles point absorption via a strength hierarchy: period > dash > semicolon > colon > comma, with adjacent points absorbing based on precedence.
• Applies quote transposition rules to move commas and periods inside or outside quotation marks based on American typographic conventions.
• Uses default feature insertion (e.g., commas for descriptive attributes) and applies graphic rendering rules to produce final DGraphR, including parentheses and formatting.

Experimental results

Research questions

• RQ1 How can punctuation phenomena such as absorption and transposition be formally modeled within a generation framework?
• RQ2 To what extent can Nunberg’s (1990) informal generalizations about punctuation be operationalized in a computational NLG system?
• RQ3 How can the order of rule application be minimized or avoided through structural design in the realization pipeline?
• RQ4 Can punctuation and formatting be uniformly treated as syntactic and morphological features rather than post-processing conventions?
• RQ5 How can complex cases involving nested parentheses, quotes, and multiple punctuation marks be consistently realized?

Key findings

• The proposed model successfully handles complex punctuation cases, such as (5.11) with nested parentheses and dashes, by applying absorption and transposition rules in a structured, rule-based manner.
• The model demonstrates that comma absorption can be modeled via a strength hierarchy (period > dash > colon > semicolon > comma), with adjacency as the sole determinant, independent of scope or word order.
• The system correctly realizes quote transposition, as in (5.22) and (5.23), where a comma after a quoted clause is suppressed when a dash is present, avoiding ungrammatical outputs.
• The use of feature structures (e.g., [btw:comma], [btw:dash]) allows for precise control over punctuation placement, with default rules triggering commas for descriptive attributes unless overridden.
• The worked example (9) shows a complete, correct realization path from DSyntR to DGraphR, including proper handling of embedded quotes, parentheses, italics, and punctuation absorption.
• The model avoids arbitrary rule ordering by embedding the strength hierarchy directly into the feature structure and rendering pipeline, reducing reliance on external sequencing.