Natural Language Processing
Natural language processing is the area of artificial intelligence concerned with enabling computers to analyze, understand, and generate human language in text or speech.
Definition
Natural language processing is the study and engineering of methods that let computers map between human language and structured representations of its form and meaning, supporting tasks from parsing and translation to extraction and generation.
Scope
This area covers the computational treatment of human language across its levels of structure: morphology and syntax (parsing), semantics and meaning representation, discourse, and applications such as machine translation and information extraction. It treats the formal models of language (grammars, logical and distributional meaning representations) and the tasks of analyzing and producing language. The general statistical and neural learning methods that train modern language models are part of the machine-learning subfield; this area emphasizes the linguistic structure, tasks, and representations specific to language.
Sub-topics
Core questions
- How is the grammatical structure of a sentence recovered from a sequence of words?
- How can the meaning of words, sentences, and discourse be represented computationally?
- How is ambiguity, pervasive at every level of language, resolved using context?
- How are language-understanding capabilities turned into applications such as translation and extraction?
Key concepts
- morphology and tokenization
- syntax and parsing
- semantics and meaning representation
- ambiguity and disambiguation
- discourse and pragmatics
- language models
- machine translation
- information extraction
Key theories
- Levels of linguistic analysis
- Language is analyzed at distinct but interacting levels—phonology, morphology, syntax, semantics, pragmatics, and discourse—and NLP systems are organized around recovering structure and meaning at these levels.
- Grammars and parsing
- Formal grammars, especially context-free and richer formalisms, model the syntactic structure of language, and parsing algorithms recover that structure, providing a backbone for meaning analysis.
- Statistical and distributional language modeling
- Treating language probabilistically—modeling the likelihood of word sequences and representing word meaning by distributional context—gave NLP robustness to ambiguity and variation and became the dominant paradigm.
Clinical relevance
Natural language processing powers search engines, machine translation, question answering and chat systems, speech recognition and dialogue, sentiment analysis, and the extraction of structured information from text in domains such as biomedicine and law, making it one of the most visibly deployed areas of AI.
History
NLP began with 1950s machine translation and the symbolic systems of the 1960s-70s, such as Winograd's SHRDLU. Statistical methods rose to prominence from the late 1980s, consolidated in texts such as Manning and Schütze (1999), and neural and large-scale language-model methods later transformed the field; its tasks and linguistic foundations remain a standard part of AI.
Debates
- Symbolic vs. statistical and neural approaches
- NLP has long oscillated between hand-built symbolic grammars and rules and data-driven statistical or neural models; the statistical turn and later neural methods came to dominate for robustness, though questions of interpretability and incorporating linguistic structure persist.
Key figures
- Daniel Jurafsky
- James H. Martin
- Christopher D. Manning
- Terry Winograd
- Karen Spärck Jones
Related topics
Seminal works
- winograd1972
- manning1999
- jurafsky2023
Frequently asked questions
- What is the difference between natural language processing and computational linguistics?
- The terms overlap heavily. Computational linguistics emphasizes using computation to understand and model human language as a scientific phenomenon, while natural language processing emphasizes engineering systems that perform useful language tasks. In practice the same models and methods serve both goals.
- Why is ambiguity such a central problem in NLP?
- Human language is ambiguous at every level: words have multiple senses, sentences have multiple parses, and references can be unclear. Much of NLP is about using context and probabilistic or learned models to choose the interpretation a human would, which is what makes the field difficult.