Data-to-Text Generation
What You'll Learn
You'll learn how neural models transform structured data—like database tables, spreadsheets, or knowledge graphs—into fluent, human-readable text descriptions.
The Core Concept
Imagine you have a sports statistics table showing player names, points scored, and game dates. A human could write: "LeBron James scored 35 points in last night's game against the Lakers." Data-to-Text Generation teaches models to do exactly that—convert structured, machine-readable information into natural language narratives.
Unlike typical text generation where the model continues from text input, here the input is structured data: rows and columns, JSON objects, knowledge graph triples (subject-predicate-object), or SQL query results. The model must understand this structure, extract relevant information, and compose coherent sentences that accurately reflect the data.
How It Works
The process typically involves:
- Encoding the Structure: The model receives structured input (e.g., a table linearized as tokens or graph representations)
- Understanding Relationships: It identifies what's important—which fields relate to each other, what's worth mentioning
- Generating Text: Using decoder architectures (like those you've learned in summarization and translation), it produces fluent descriptions
Common applications include:
- Weather reports from meteorological data
- Financial summaries from earnings tables
- Product descriptions from specification databases
- Sports recaps from game statistics
The challenge lies in faithfulness—ensuring the generated text accurately reflects the source data without hallucinating facts—and fluency—making it read naturally rather than like a robotic list.
Key Differences from Other Generation Tasks
While abstractive summarization transforms text-to-text, data-to-text must bridge the gap between formal structure and natural language, requiring models to "verbalize" numbers, dates, and relational information meaningfully.
Key Takeaway: Data-to-Text Generation converts structured data formats into natural language descriptions, enabling automated report writing and making databases accessible through human-readable narratives.