AI Glossary / Natural language and speech / Beginner
Word sense
Definition
Definition: A word sense is a particular meaning associated with a word. In computational language work, it often means a distinct meaning category in a chosen dictionary or other inventory, used to interpret that word in context.
What it means
What it means: The spelling crane can describe a bird or a machine used for lifting. These are different senses even though the written form is identical. Recognizing the spelling is therefore not the same as understanding the occurrence.
A sense belongs to a word as used with a meaning, rather than to every occurrence of its letters. Inflected forms can share a sense, while two occurrences of the same form can differ. The boundary between senses is also a modeling decision: one inventory may separate closely related uses that another groups together. A useful system states which distinctions it needs instead of assuming one universally correct number of meanings.
How it works
How it works: A sense inventory provides meaning labels, explanatory definitions and sometimes examples or relationships to other meanings. An application associates a word occurrence with an appropriate entry using its sentence, surrounding document and task context.
Designers first decide the required level of detail. A wildlife-versus-construction distinction may suffice for broad search, while a specialized translation task may need finer distinctions. They then write annotation guidance and test whether different readers can apply it consistently. A contextual embedding can represent differences among occurrences without itself providing a named dictionary sense. Turning that representation into an inventory label requires an additional mapping or decision.
Three practical examples
Three practical examples:
1. A document search tool distinguishes construction articles mentioning a crane from birdwatching reports. Matching the same letters remains useful for retrieval, but sense-aware ranking can separate the intended topics.
2. A translation assistant examines the context around seal before deciding whether the text concerns an animal, a closure or an official mark. The distinction is about meaning, not merely the word's part of speech.
3. A product-support team gives charge separate internal labels for an electrical reading and an amount billed. It documents these task-specific categories so that annotators do not silently use incompatible dictionary distinctions.
Think of it this way
Think of it: A word sense resembles a labeled compartment in a filing system for meanings. The written word may point to several compartments, and context helps choose the appropriate one. The analogy has limits: language develops new uses, meanings can overlap, and a filing scheme is a practical representation rather than a perfect map of human understanding.
Worked example
Worked example: Create a deliberately small inventory for crane: B means the bird and M means the lifting machine. Label four invented sentences: “The crane nested beside the marsh” is B; “The crane lifted the steel beam” is M; “A crane fed near the reeds” is B; and “The crane operator checked the load” is M.
There are four occurrences of one written word form and two represented senses. A rule that assigns M to every occurrence gets two of four labels correct, or 50% accuracy. That result does not mean crane has exactly two meanings in every dictionary: the exercise deliberately supplies a restricted inventory and only noun examples.
Now add “The crane was nearby.” Without further context, either label may fit. Do not count an arbitrary guess as a verified interpretation. An annotation process can record ambiguity, request more context or apply an explicitly agreed convention. Its evaluation must say how such cases are handled.
Illustrative case study
Illustrative case study: Fictional Reed Archive organizes maintenance documents and nature reports in one search portal. Users searching crane receive a mixed result list. Editors create a small meaning inventory and annotate a sample containing both domains.
During review, they find that several short headings lack enough context to classify reliably. The team retains an uncertain category for review instead of forcing those headings into the larger class. It also keeps the original text alongside the labels so that future editors can reconsider decisions. This invented case illustrates inventory design and ambiguity management; it does not claim a measured improvement from adding sense labels.
Common misconceptions and mistakes
Common misconceptions and mistakes: A word sense is not a separate spelling, token or pronunciation by definition. Part of speech helps interpretation but does not resolve all ambiguity: both a bird and a lifting machine can be nouns. A dictionary entry is not necessarily one sense, and different dictionaries need not divide meanings identically. Related words can share a topic without being interchangeable in the intended sense. Finally, a model's plausible explanation is not proof that the text provides enough evidence for a unique interpretation.
Business use
Business use: Sense distinctions can support search, translation, routing and terminology management. Begin with the decisions the application must make, then choose an inventory detailed enough for those decisions. Save inventory versions and annotation guidance with the data. Include ambiguous examples and uncommon meanings in evaluation rather than measuring only the dominant use. Where a mistaken interpretation affects a customer action, preserve context and provide a review path instead of hiding uncertainty behind a single confident label.
People also ask
People also ask: Is a sense the same as a word?
No. One word can have several senses, and a sense describes a particular meaning rather than just its written form.
Does every dictionary list the same senses?
No. Dictionaries and computational inventories can differ in purpose, coverage and how finely they separate related uses.
Does an embedding automatically identify a dictionary sense?
No. It can represent contextual information, but mapping that representation to an explicit inventory requires a defined method.
Check your understanding
Check your understanding: Two reports use crane, one beside “nest” and another beside “lifting capacity.” Must the matching spelling imply matching meanings?
Answer: No. Those contexts support different interpretations. A sense-aware system should evaluate the occurrences separately under its chosen inventory.
Related terms
- Word sense disambiguation
- Homonymy
- Lemma
- Contextual embeddings
- polysemy
- word form
- synset
- semantic relation
- contextual embedding
- sense inventory
Learn next
choose a word with two clearly different uses. Write context-rich examples and one ambiguous example, define your labels, and ask another reader to apply them. Discuss disagreements before adding finer distinctions or treating the labels as unquestionable ground truth.
More in Natural language and speech
- Wikification - Definition: Wikification is the process of identifying relevant terms or mentions in a text and linking them to the appropriate Wikipedia articles.
- Word error rate - Word error rate, or WER, measures the word-level differences between a speech-recognition transcript and a reference transcript.
- Word list - A word list is an organized collection of words selected for a stated purpose.
- Word-sense induction - Word-sense induction is the automatic discovery of different uses or senses of a word from the contexts in which it occurs.
- Word shape - Definition: Word shape is a representation of a token's character pattern, such as uppercase letters, lowercase letters, digits and punctuation.
- Wordpiece - Definition: WordPiece is a subword tokenization method that represents text using a fixed vocabulary containing whole words and smaller word pieces.
Sources
Sources: Princeton WordNet, official explanation of senses, synsets and lexical relations: https://wordnet.princeton.edu/ . Jurafsky and Martin, Speech and Language Processing, Appendix I, Word Senses and WordNet: https://web.stanford.edu/~jurafsky/slp3/I.pdf . The restricted crane inventory, calculations and Reed Archive case are original illustrations, not reported corpus results or a reproduction of an official dictionary entry.
Previous: Word list Next: Word sense disambiguation