• Talk: Limits and Applications of Natural Language Processing
• NLP
  • subfield of AI
  • siri, amazon, grammarly
• sometimes, they can very much fail and create big isues
• to improve NLP systems, we must understand where they fail
• Natural Language Inference
  • used to test our systems ability to understand inference
• word embeddings. need to convert words in a way that a computer can understand
  • have a unique representation of each word
  • use context
• co-occurence matrix
• Word2Vec: Skip-Gram
  • give neural network a single word
• word embeddings preserve meaning
• Recasting
  • leverge existing data sets...
• machine translation....lots of work...lots of hype around it...
• anaphora resolution in MT
• paraphrastic inference in MT
• NLP → HLT
• ML ↔ HLT → CL
  • interested in studying language
• use methods from NLP to study language
• algorithm that can learn from data
• goal of ML: algorithmic contribution
• AI ↔ HLT → CTA (computational text analysis)
  • analyzing large amounts of tects
• what is computational text analysis?