Where does chunking collide?

The grouping of separate items into a smaller number of larger, meaningful units, so that more can be held in the narrow span of short-term memory. Miller's 1956 The magical number seven, plus or minus two argued that immediate memory holds only about seven units at once, whatever each unit carries, so recoding raw items into richer chunks raises how much a person can retain. Chase and Simon's 1973 Perception in chess showed that expert players recall a board in a few large, familiar chunks rather than many single pieces, tying the chunk to knowledge already learned. The referent is a unit of memory inside a person's head, measured by span and recall, not a slice of a data file, a segment of a lesson, or a phrase marked in a sentence.[1][2]

Unit

A meaningful unit held in short-term memory

Recovery means

Recall leans on patterns already in long-term memory; items that match no pattern cannot be chunked and are lost faster

Measured by

Tested by memory span and recall tasks, comparing what is retained when material can or cannot be grouped

In computer systems, cutting a file or data stream into pieces so that each piece can be stored, addressed, deduplicated, or transferred on its own. Muthitacharoen, Chen, and Mazieres's 2001 A low-bandwidth network file system split files at boundaries chosen by the data's own content with a rolling hash, so that inserting bytes shifts only the nearby chunks and unchanged chunks can be recognised and skipped over the network. Chunks can instead be cut at fixed sizes; either way the goal is to break one large object into parts that are handled independently. The referent is a run of bytes bounded by an algorithm, counted in kilobytes and boundaries, and it shares nothing with a unit of memory, a lesson segment, or a grammatical phrase beyond the word.[3]

Unit

A segment of a byte stream, cut by fixed size or by content

Recovery means

Not applicable

Measured by

Judged by chunk size distribution, boundary stability under edits, and how much duplicate data is avoided

In learning and instruction, grouping new material into meaningful segments so that it fits within a learner's limited working memory and can be built into larger units with practice. Gobet and colleagues' 2001 Chunking mechanisms in human learning reviewed how repeated exposure lets people form and store chunks that make later material easier to take in, treating chunking as a general mechanism behind skill and expertise. Teachers and instructional designers use the same idea when they break a lesson into manageable parts to keep the load on working memory within bounds. This meaning is a close relative of the memory chunk rather than a pure accident of spelling, but it is studied as a route to learning, not as a slice of stored data or a parsed phrase.[4]

Unit

A meaningful segment of material to be learned

Recovery means

Well-formed chunks let a learner rebuild material from a few cues; poorly grouped material must be held item by item

Measured by

Judged by learning and transfer outcomes, and by measures of cognitive load during instruction

In natural language processing, chunking, also called shallow parsing, divides a sentence into non-overlapping spans such as noun phrases and verb phrases without building a full parse tree. Tjong Kim Sang and Buchholz's 2000 Introduction to the CoNLL-2000 shared task: Chunking defined the task of labelling each word with the phrase chunk it belongs to and set a shared benchmark for scoring it. A chunk here is a contiguous group of words picked out by their grammatical role. The referent is a syntactic span in text, scored against hand-annotated data, not a unit of memory, a block of stored bytes, or a lesson broken into parts.[5]

Unit

A non-overlapping phrase span within a sentence

Recovery means

Not applicable

Measured by

Scored by precision, recall, and F-measure against hand-annotated chunk boundaries