History

• Symbolic (象征) NLP (1950s – early 1990s)
  • 基于明确的规则和逻辑推理
  • 可解释性强，能理解复杂语言逻辑
  • 很难写全所有规则，应对不了语言中的灵活表达
• Statistical (统计) NLP (1990s–2010s)
  • 依赖大量数据和概率计算
  • 适合处理大规模数据
  • 需要大量训练数据，结果不一定能解释清楚
• Neural (神经) NLP (present)
  • 模拟人脑工作原理，通过层层计算“自动学习”语言
  • 适合复杂任务，比如翻译、聊天、生成文章
  • 计算量大，训练时间长，结果不容易解释

Common tasks

• Text and speech processing
  • Optical (光学) character recognition (OCR)
  • Speech recognition
  • Speech segmentation
  • Text-to-speech
  • Word segmentation (Tokenization) 单词分割（令牌化）
• Morphological (形态) analysis
  • Lemmatization
  • Morphological segmentation
  • Part-of-speech tagging
  • Stemming
• Syntactic(句法) analysis
  • Grammar induction
  • Sentence breaking (also known as “sentence boundary disambiguation”)
  • Parsing
• Lexical semantics (of individual words in context)
  • Lexical semantics
  • Distributional semantics
  • Named entity recognition (NER)
  • Sentiment (情感) analysis (see also Multimodal sentiment analysis)
  • Terminology (术语) extraction
  • Word-sense disambiguation (WSD)
  • Entity linking
• Relational semantics (semantics of individual sentences)
  • Relationship extraction
  • Semantic parsing
  • Semantic role labelling (see also implicit semantic role labelling below)
• Discourse (semantics beyond individual sentences)
  • Coreference resolution
  • Discourse analysis
  • Implicit semantic role labelling
  • Recognizing textual entailment
  • Topic segmentation and recognition
  • Argument mining
• Higher-level NLP applications
  • Automatic summarization (text summarization)
  • Grammatical error correction
  • Logic translation
  • Machine translation (MT)
  • Natural-language understanding (NLU)
  • Natural-language generation (NLG):
  • Book generation
  • Document AI
  • Dialogue management
  • Question answering
  • Text-to-image generation
  • Text-to-scene generation
  • Text-to-video
  • llm

References