Skip to content

Latest commit

 

History

History
56 lines (49 loc) · 2.05 KB

Natural_Language_Processing.md

File metadata and controls

56 lines (49 loc) · 2.05 KB
Raw

WWDC 2017

Table of Contents

Natural Language Processing (NLP) - Wednesday

Session video and resources: https://developer.apple.com/videos/play/wwdc2017/208/

  • Typed, recognized handwriting, transcribed speech

NLP APIs

  • Natural language text -> Language identification
  • Natural language text -> Tokenization -> Tokenize text
  • Natural language text -> Part of speech -> Assign parts of speech
  • Natural language text -> Lemmatization -> Lemmatize text
  • Natural language text -> Named entity recognition

NSLinguisticTagger

  • Used for segmenting and tagging text
  • Omitting whitespaces and punctuation is also possible
  • Pass in schemas for options like lemmatization, language identification etc.

Tagging Units

  • public enum NSLinguisticTaggerUnit
    • word
    • sentence
    • paragraph
    • document
  • dominantLanguage
  • Improved performance
  • Higher accuracy
  • Additional language support

Use Case

  • A simple photo app with tagged descriptions
    • A query for an example word like hike will return related results via NLP. Rather than just searching for the exact word hike
    • Recognizes dominant language and analyses through lemmatization
  • A social media app merging all major social media providers
    • Organize feeds by people, organization and location using NLP

Why should you use the NLP API's

  • Homogeneous text processing
    • Consistent UX and results across all platforms and devices
  • Privacy
    • On-device machine learning
  • Performance
    • Multi-threaded
    • 50K tokens/sec in a single thread, 80K tokens/sec in a multi-thread
  • Language support
    • 29 scripts, 52 languages for language identification
    • Tokenization for all iOS/macOS languages