Is a dictionary the same as a vocabulary list?

A list may contain only expressions; a dictionary records additional information.

Is WordNet a conventional alphabetical dictionary?

It is a lexical database organized around senses and their relationships.

Can a dictionary work with a language model?

Yes. An application can provide retrieved lexical information, although the model's use of it still needs evaluation.

AI Glossary / Graphs and symbolic AI / Advanced

Dictionary

Definition

In the natural-language-processing sense used here, a dictionary is a structured resource that records words or expressions and information about them. Entries can distinguish meanings, grammatical roles, definitions, and relationships to other expressions. This lexical resource is different from the programming data structure also called a dictionary, which associates keys with values.

What it means

A dictionary gives a language-processing system an explicit inventory of lexical knowledge. Instead of treating a word only as a sequence of characters, the system can consult recorded information about its possible meanings and uses. Human editors often build or review such resources, although construction methods vary.
A word form and a meaning are not the same thing. One spelling can have several meanings, while different expressions can refer to a similar concept. A useful lexical representation keeps these distinctions visible. Looking up an entry supplies candidate information; deciding which meaning applies in a particular sentence remains another task.

How it works

An application first identifies the expression to look up. It may need to handle capitalization, inflected forms, or expressions containing several words. The lookup then returns entries associated with the expression, potentially separated by part of speech and meaning.
WordNet illustrates a richer lexical organization. It groups expressions by particular senses and records relationships among the resulting concepts. A relation saying that one concept is a kind of another differs from a relation saying that it is part of another. Keeping relation types explicit prevents an application from treating every connection as interchangeable meaning.
A downstream component uses the retrieved information for its task. It might offer definitions, expand a search query, or compare candidate interpretations. The resource's coverage and the application's selection rules both affect the result. A successful lookup does not demonstrate that the surrounding sentence has been understood.

Three practical examples

Search assistance: An internal search tool uses approved alternative terms to retrieve documents written with different wording. Expansion should follow the relevant meaning; adding every alternative associated with an ambiguous word can introduce unrelated results.
Terminology support: A drafting assistant consults a domain dictionary for preferred expressions and definitions. Editors can inspect the underlying entries when a suggestion seems inappropriate. Specialized meanings may need an organization-specific extension to a general resource.
Text analysis: A document-processing system retrieves candidate meanings for unfamiliar expressions before assigning a topic. The dictionary provides explicit lexical information, while context or human review helps determine whether a candidate actually fits the passage.

Think of it this way

Imagine a reference cabinet with separate cards for different meanings of the same expression. Related cards have labeled links explaining their relationship. Finding the right drawer reveals possible interpretations; choosing the card that fits a sentence still requires attention to that sentence.

Worked example

Create an invented dictionary for a small document collection. It records two noun senses of “crane”: a lifting machine and a long-necked bird. These are deliberately simplified entries, not a reproduction of a particular published dictionary.
Suppose 20 documents contain that spelling. Human review establishes that 12 discuss lifting equipment and 8 discuss birds. A lookup that returns every document containing the word retrieves all 20. For a user seeking lifting equipment, 12 of the 20 results are relevant: precision is 12/20 = 60%, while recall for the 12 relevant documents is 12/12 = 100%.
Now add a context-selection component that returns 11 documents: 10 about equipment and 1 about birds. Precision becomes 10/11, approximately 90.9%, and recall becomes 10/12, approximately 83.3%.
The dictionary made the two candidate senses explicit, but the selection component produced the improved precision and reduced recall in this invented evaluation. It would be misleading to attribute those performance figures to dictionary lookup alone.
The same example also distinguishes coverage from interpretation. Both senses can be present in the resource even when the application selects the wrong one. Conversely, no selection rule can retrieve an absent specialized definition directly from that dictionary; another source or an extension would be needed.

Illustrative case study

Fictional Willow Manuals introduces terminology assistance for equipment documentation. Reviewers discover that a general lexical resource covers ordinary meanings but misses several locally defined component names. Some suggestions also use a related concept where an exact replacement is required.
The team creates a reviewed extension containing local definitions, usage notes, and approved alternatives. It labels relationship types and keeps the source and version of each addition. Tests include ambiguous words, inflected forms, and expressions whose meaning depends on the surrounding procedure.
The team evaluates the assistance against the documentation task rather than counting dictionary matches as correct interpretations. This is a fictional workflow illustrating resource maintenance and contextual evaluation, not a deployment result.

Common misconceptions and mistakes

“Each spelling has one dictionary meaning.” Many expressions have several senses.
“Related words can always replace one another.” Their relationship and context matter.
“A lookup resolves the sentence's meaning.” It usually supplies information for further interpretation.
“Missing from the dictionary means invalid language.” The resource may lack the expression or intended sense.
“A Python dictionary is automatically a linguistic dictionary.” A key-value container can store lexical entries, but the container itself supplies no lexical knowledge.

Business use

Choose a lexical resource that fits the language, domain, and intended task. Check usage rights, coverage, update practices, and whether the required information is actually represented. A large entry count alone does not establish suitability for specialist documents.
Maintain reviewed extensions when local terminology matters. Record which resource version supports an output so that changes can be investigated. Evaluate useful task outcomes, such as finding the right document or suggesting an appropriate expression, separately from raw lookup coverage.

Check your understanding

Why separate the two crane senses?

Answer: They refer to different concepts despite sharing a spelling.

Why is the second retrieval result not perfect?

Answer: It includes one irrelevant document and misses two relevant ones.

What does a successful lookup prove?

Answer: That matching information was found, not that the correct contextual meaning was selected.

Related terms

Lexicon
Word sense
Lemma
Word sense disambiguation
synset
lexical database
query expansion
terminology management

Learn next

Build a small domain dictionary with separate sense identifiers and labeled relationships. Test it on ambiguous sentences, recording missing entries separately from incorrect sense selections and inappropriate synonym substitutions.

Sources

AIMA glossary, dictionary entry and intended lexical sense: https://github.com/aimacode/aima-glossary/blob/master/glossary.md Princeton WordNet, lexical database overview: https://wordnet.princeton.edu/ Princeton WordNet, terminology and database organization: https://wordnet.princeton.edu/documentation/wngloss7wn Python tutorial, dictionaries as key-value data structures: https://docs.python.org/3/tutorial/datastructures.html#dictionaries

Previous: Dichromatic model Next: Dictionary-based machine translation