Core Chunking Terms

Chunking: The process of grouping words in a sentence into syntactic phrases (chunks), such as noun phrases (NP), verb phrases (VP), and prepositional phrases (PP), without building a full parse tree.

Chunk: A non-overlapping, non-recursive group of words that form a meaningful unit. Example: "the current account deficit" (NP), "will narrow" (VP).

IOB Tagging: A scheme for marking chunk boundaries:

• B- (Beginning): Marks the first word of a chunk
• I- (Inside): Marks subsequent words in the same chunk
• O (Outside): Marks words not part of any chunk

Noun Phrase (NP): A chunk containing a noun and its modifiers. Example: "the red ball" Verb Phrase (VP): A chunk containing a verb and its auxiliaries or complements. Example: "is running fast" Prepositional Phrase (PP): A chunk starting with a preposition and its object. Example: "in the park"

Machine Learning Terms

Hidden Markov Model (HMM): A probabilistic model for sequence labeling, used to predict chunk tags based on word and tag probabilities. Conditional Random Field (CRF): A discriminative model for structured prediction, allowing richer feature sets for chunking.

Feature: A property or attribute of a word or its context (e.g., lexicon, POS tag) used by the model to make predictions. Training Corpus: A labeled dataset used to train the chunker.

Analysis Terms

Chunk Boundary: The point in a sentence where one chunk ends and another begins. Gold Standard Annotation: The correct chunking of a sentence, used for evaluation. Chunking Accuracy: The percentage of correctly identified chunks compared to the gold standard. Shallow Parsing: Another term for chunking, focusing on phrase-level analysis rather than full syntactic parsing.