Introduction to Natural Language Processing

Study Notes

NLP is a field that combines computer science, artificial intelligence, linguistics, and data science to enable computers to understand and process human language.
NLP uses machine learning and deep learning techniques to perform tasks like language translation, question answering, and text summarization.
NLP takes unstructured data and transforms it into a structured format.

NLP involves several subcategories: Natural Language Understanding (NLU) and Natural Language Generation (NLG).
NLU focuses on comprehension, syntactic and semantic analysis to determine sentence meaning, and establishing an ontology (data structure to specify relationships between words and phrases).
NLG focuses on computer writing, producing human language text based on input data.
NLG can generate summaries from documents while maintaining integrity, or use templates to generate text.

NLP uses methods from various disciplines to enable computers to understand human language.
NLP uses machine learning and deep learning to execute tasks.
NLP converts unstructured data to a structured format, like using word patterns (ex: tokenization, stemming, lemmatization) to find the root word form.

Text-based applications include processing written text (books, newspapers, reports, emails).
Dialogue-based applications involve human-machine communication, often using spoken language, databases of texts, or finding and extracting information.

There are two categories for evaluating these systems:
- Black Box Evaluation: assess the system's performance without understanding its internal workings. It is evaluated based on whether it performs the task it is designed to do correctly (eg: answering questions based on a database).
- Glass Box Evaluation: evaluates the system's internal structure and examines its different components. It requires consensus on what the internal components of NL systems should be.

NLP systems must apply knowledge about language structure to determine the meanings of words and sentences.
Human general world knowledge and reasoning abilities play a role in comprehending and producing language in conversations or addressing questions

NLU Systems use processes that map from one representation to another (eg: mapping a sentence to its syntactic and logical form).
The parser uses knowledge about words (lexicon) and grammar rules to assign syntactic structure and logical form to an input sentence.

The logical form encodes semantic relationships between words and phrases to understand sentence meaning.
Identifying semantic relationships enables understanding different possible meanings and eliminates impossible ones.
Important to consider the combination of word meanings to construct coherent sentence meanings and reduce the number of possible interpretations of a sentence.

Final representation is a general knowledge representation (KR).
Understanding the context of a sentence and mapping it to expressions within the knowledge base of a specific application allows the system to complete relevant tasks.
The KR is used to represent and reason about a domain; For instance, a question in a question-answering system might be mapped to a query on a database.

Systems need to consider how words are put together in sentences, the relationships between words and phrases, and what roles each word plays..
Different ways of organizing NLP components are possible(ex: performing syntactic processing and semantic interpretation in turn).