How to build resilient NLP applications

2020-11-24 · Source: Explosion · Developer tools and consulting for AI, Machine Learning and NLP - Explosion.ai · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Advanced, extended

Summary

Explosion founders Ines Montani and Matt Honnibal discussed the spaCY 3.0 release and the challenges of building production-grade NLP applications. spaCY 3.0 introduces Transformer-based pipelines, achieving near state-of-the-art accuracies with efficient multitask learning, and a new configuration system. This system aims to manage the inherent complexity of machine learning by providing a single, customizable configuration file, moving away from over-abstraction. They emphasize that successful NLP projects require domain expertise, iterative improvements based on user insights, and a focus on designing solvable problems rather than solely chasing benchmark scores or "magic models." The discussion also touched on the evolution of ML into mainstream software engineering, the importance of open-source tools like spaCY for processing large text volumes, and Explosion's annotation tools, Prodigy and Prodigy Teams, which support data development workflows for larger teams. The future of ML is seen as consolidation and spreading expertise, not just breakthroughs.

Key takeaway

For Machine Learning Engineers building production NLP systems, spaCY 3.0 offers a robust framework to manage complexity. You should utilize its new configuration system to define custom neural networks and Transformer-based pipelines, gaining granular control over model behavior. Prioritize integrating domain expertise and iterative data development using tools like Prodigy, rather than solely optimizing for generic benchmarks. This approach ensures your applications deliver real business value and adapt effectively to specific use cases.

Key insights

Production-grade NLP requires embracing inherent complexity, providing granular control, and integrating domain expertise for real-world value.

Principles

• Machine learning is now a foundational piece of software engineering.
• Successful NLP projects prioritize domain expertise over generic metrics.
• Over-abstracting ML complexity leads to less programmable systems.

Method

Define entire NLP pipelines and models in a single, bottom-up resolved configuration file, managing settings and creating objects for custom neural networks.

In practice

• Use spaCY 3.0's Transformer pipelines for state-of-the-art accuracy.
• Integrate annotation tools like Prodigy for data development and team collaboration.

Topics

• NLP Libraries
• spaCY 3.0
• Production-Grade AI
• Machine Learning Workflows
• Data Annotation
• Transformer Models
• Open-Source Development

Best for: AI Engineer, NLP Engineer, Machine Learning Engineer, Software Engineer

Related on AIssential

• Evolution of spaCy — Industry-grade NLP requires opinionated, fast, and extensible tools that prioritize developer engagement with data.
• Building new NLP solutions with spaCy and Prodigy — NLP project success hinges on iterative design, data annotation, and model integration, not just model architecture.
• Getting My Hands Dirty with Python and NLP — Building a practical NLP project is an effective way to learn Python libraries and data structures.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Explosion · Developer tools and consulting for AI, Machine Learning and NLP - Explosion.ai.