D2S – Data-to-Speech & Advanced NLG System | NLG-Wiki

D2S – An Overview of the Data-to-Speech NLG System

What is D2S?

D2S (Data-to-Speech) is a natural language generation (NLG) system engineered to transform structured data into coherent text and optionally speech output. It serves as a reference case for systems that bridge data sources, linguistic structures, and spoken output.

Architecture and Design

D2S uses a template-driven generation approach, organised around syntactically enriched templates (resembling Tree-Adjoining Grammar structures) that map data representations into fluent text.

Key features include:

• Input type: Structured event or fact records (for example: goal events in sport reports)

• Template library: Trees or templates annotated with conditions, slot-fillers and discourse topics.

• Speech module (optional): Text output may be passed to a speech generation module for spoken realisation.

• Generation strategy: While template-based, D2S integrates mechanisms for referring expression generation and context modelling, improving linguistic robustness.

Application Domains

Though originally developed in academic research, D2S has been used in domains such as:

• Automated sports commentary and report generation (e.g., goal reports)

• Multilingual data-to-speech prototypes for assistive technologies

• Domain-specific generative systems where structured databases feed natural text output

Why D2S Matters

D2S retains relevance in NLG research for several reasons:

• It demonstrates how template-based systems can offer strong linguistic structure while remaining practically deployable.

• It shows a clear pipeline from data to text to speech, highlighting considerations of latency, discourse, and prosody.

• It offers a documented case for comparison with newer neural or hybrid NLG systems, particularly in terms of modularity, input control, and domain-specific adaptation.

How to Explore D2S on NLG-Wiki

1. Locate the “Systems” section and search for D2S.

2. Review metadata: architecture type (template-based), input/output, domain examples, licensing details (if available).

3. Consult the linked research papers (e.g., Theune et al.) for deeper insight into D2S’s technical design.

4. Compare D2S with other NLG systems listed (e.g., Wotan, OntoGeneration) to assess strengths and trade-offs.

FAQ – D2S NLG System

What is the primary function of D2S?
It converts structured data into coherent natural language text and optionally speech, supporting applications that require accurate verbalisation of data.

Is D2S rule-based or statistical?
Primarily rule/template-based, though it incorporates linguistic modules (e.g., referring expression generation) to enhance quality.

Can D2S be used in multiple languages?
Yes, in principle: the architecture is designed to support multilingual output by adapting lexicons and grammars, though domain-specific use often limits language scope.

Where can I find more information about D2S?
The NLG-Wiki entry provides references, and the original research publications (e.g., Theune et al.) give in-depth descriptions.