VIE-GEN – Sentence Generation from a Semantic Network for German (NLG System)
Table of Contents
VIE-GEN – A Generator for German Texts
VIE-GEN is a natural language generation system developed in the 1980s as a generator for German texts. It forms the generation component of the German language dialogue system VIE-LANG, where it produces German sentences from a structured semantic representation. The input to VIE-GEN comes from the episodic layer of a semantic network called SEMNET, and the system’s task is to turn this conceptual structure into fluent, grammatically correct German output.
In terms of architecture, VIE-GEN is a sentence generation system based on a semantic network, designed to reflect the classical distinction between deciding what to say and how to say it. The dialogue and knowledge components determine the content, while VIE-GEN is responsible for linguistic realization. To handle this, the generator works in two main phases: a verbalization phase and a realization phase. During verbalization, the system selects and structures the information to be expressed, mapping semantic relations onto abstract sentence plans. In the realization phase, these plans are transformed into concrete German sentences with appropriate word order, morphology, and syntactic structure.
VIE-GEN was explicitly tailored to the idiosyncrasies of German grammar, which makes it an important early example of a language-specific NLG system. It can generate different word orders to reflect the relatively free constituent order in German, and it distinguishes carefully between main clauses and subordinate clauses. The system also covers inflectional morphology, producing the correct case, number, person, tense, and agreement patterns required by German syntax. This detailed handling of word order and inflection was essential for generating text that sounds natural to native speakers rather than mechanically assembled.
Beyond single sentences, VIE-GEN includes mechanisms for building coherent multi-sentence output. It supports features such as anaphora (for example, choosing pronouns instead of repeating full noun phrases) and gapping or ellipsis, where certain elements are omitted because they are recoverable from context. These discourse-level capabilities allow the generator to produce short stretches of text that are more cohesive and human-like, rather than isolated sentences with no connection to each other.
Historically, VIE-GEN is part of the early generation of rule-based NLG systems documented in the book Natural Language Generation Systems. It demonstrates how a semantic network, a dialogue system, and a language-specific generator can be integrated into a complete natural language dialogue architecture. For researchers and students of NLG, VIE-GEN serves as a case study in semantic-to-surface generation for German, illustrating issues such as lexical choice, word order variation, sentence structuring, and discourse phenomena in a pre-neural, symbolic setting.
Although more recent work in natural language generation relies heavily on statistical and neural methods, VIE-GEN remains relevant as a classic German NLG system that shows how carefully designed rules, semantic representations, and modular generation phases can yield high-quality text in a morphologically rich language. It is often cited in surveys and overviews of natural language generation as an example of a semantic-network-based generator and continues to be referenced in discussions of multilingual sentence generation, lexical semantics, and knowledge representation.