AI Glossary / Natural language and speech / Beginner
Dictionary-based machine translation
Definition
Definition: Dictionary-based machine translation translates source expressions by looking up corresponding target-language entries in a bilingual dictionary. In its simplest form, it replaces words or listed phrases with equivalents; more capable dictionary-driven systems add analysis, selection and grammatical processing.
What it means
What it means: A bilingual dictionary supplies possible correspondences, not a complete account of how to translate every sentence. A source word may have several meanings, and a target equivalent may need a different form or position. Multiword expressions can also require an entry or analysis that treats them together.
The label is broad and can be used loosely. This entry focuses on dictionary lookup as the main translation mechanism and distinguishes it from larger rule-based systems that use dictionaries alongside other components. Almost any translation workflow can consult terminology, but merely using a glossary does not make the whole system a simple word-replacement translator.
How it works
How it works: A basic system segments the input, finds matching entries and emits selected equivalents. It needs a policy for unknown words, competing entries and phrases that overlap. A more elaborate pipeline can reduce inflected words to lemmas, preserve grammatical features and use contextual rules to select a sense.
Additional transfer and generation stages can reorder elements and create appropriate target-language forms. Apertium's documentation provides a concrete example of bilingual lexical transfer inside such a broader pipeline. Its use of a dictionary should therefore not be mistaken for a claim that the complete system translates by context-free substitution alone. The surrounding linguistic resources are part of the method.
Three practical examples
Three practical examples:
1. A teaching prototype translates a tightly limited list of object labels. The input vocabulary and intended senses are controlled, making dictionary coverage easy to inspect.
2. A technical translation workflow uses an approved bilingual glossary to suggest consistent equivalents. Human review or other translation components still determine grammar and contextual meaning.
3. A rule-based language-pair project stores lemma correspondences in a bilingual lexicon. It maintains morphological and transfer resources separately so that lexical lookup is not expected to solve every structural difference.
Think of it this way
Think of it: Dictionary-based translation resembles replacing labeled parts using a correspondence catalog. The catalog helps identify candidate replacements, but assembling a working result also requires attention to how the parts fit. The analogy is limited because meaning often belongs to a whole expression rather than to independent interchangeable words.
Worked example
Worked example: Use the original teaching phrase “red houses.” A tiny English-to-Spanish dictionary contains house → casa and red → rojo. A strict surface-word lookup finds red but may fail on houses if only singular house is listed. A morphological step can recognize houses as the lemma house with a plural feature.
Lexical lookup then supplies casa and rojo as base forms. To produce the ordinary intended Spanish phrase “casas rojas,” the pipeline also needs plural formation, feminine agreement for the adjective and the appropriate adjective - noun order for this example. Merely outputting “rojo casa” loses both the number and the required agreement, while preserving the English order does not give the intended ordinary phrasing here.
This small example isolates four different responsibilities: recognizing an inflected input, choosing lexical equivalents, arranging them and generating their forms. It is not actual output from Apertium or a claim that every Spanish adjective follows a noun. A real translation needs the construction-specific rules and context.
Illustrative case study
Illustrative case study: Fictional Hazel Labels maintains bilingual descriptions for a small equipment catalog. A lookup prototype translates familiar nouns consistently but mishandles an ambiguous word used for two different components. Adding another equivalent without a selection rule makes the output less predictable.
The team records domain-specific senses and tests complete descriptions, including plural forms and multiword terms. It distinguishes missing dictionary coverage from incorrect sense selection and grammatical generation errors. This invented case illustrates resource maintenance. It does not claim that a larger dictionary alone can reach professional translation quality or that a terminology database removes the need for contextual review.
Common misconceptions and mistakes
Common misconceptions and mistakes: A dictionary entry is not always a one-to-one equivalence. A translation valid in one sense can be wrong in another. Adding every inflected form manually is not the only option; morphological analysis can relate forms to lemmas. Conversely, a correct lemma translation does not guarantee a correctly inflected sentence. Longest-phrase matching can help with listed expressions but does not solve all ambiguity. Unknown-word handling must also be explicit rather than silently treating every unmatched term as safely copyable.
Business use
Business use: Assess the intended input scope before considering a lookup-based approach. Controlled labels and terminology suggestions pose different demands from unrestricted prose. Track coverage, sense accuracy, phrase handling and grammatical output separately. Version dictionaries and record why a preferred equivalent was selected, especially when several teams maintain terminology. Review complete translated examples after resource changes because an entry can affect many contexts. Where the output is only a draft or lexical aid, present it accordingly so users do not mistake word coverage for reliable sentence translation.
People also ask
People also ask: Is dictionary-based translation always word by word?
The simplest version is, but dictionaries can include multiword expressions and can be embedded in richer processing pipelines.
Is it the same as all rule-based translation?
No. Rule-based systems may perform substantial grammatical analysis and transfer in addition to lexical lookup.
Will a complete dictionary eliminate translation errors?
No. Context, ambiguity, grammar and discourse can still change which equivalent or construction is appropriate.
Check your understanding
Check your understanding: A system correctly maps a plural source noun to a target lemma but outputs that lemma in singular form. Which distinction has it missed?
Answer: Lexical correspondence and grammatical form generation are separate. Finding the right lemma does not by itself preserve number.
Related terms
- Rule-based machine translation
- Lemma
- bilingual dictionary
- lexical transfer
- morphological analysis
- lexical selection
- structural transfer
- terminology management
Learn next
build a tiny fictional bilingual lexicon with one ambiguous word and one multiword expression. Specify the allowed input context and unknown-word policy. Then list the additional information needed to translate an inflected sentence rather than just isolated labels.
More in Natural language and speech
- Determiner - A determiner is a grammatical word associated with a noun phrase that helps express how its referent should be understood in context.
- Development set - A development set is a collection of examples used to guide choices while building a machine-learning system.
- Dialogue act - A dialogue act is the communicative function performed by an utterance or part of an utterance in a conversation.
- Digit recognition - Digit recognition is the task of identifying digits from an input signal.
- Digital divide - The digital divide is unequal access to, ability to use, or benefit from digital technologies and information.
- Digitization - Digitization is converting information from a physical or analog form into a digital representation that a computer can store and process.
Sources
Sources: Apertium, Bilingual dictionary: https://wiki.apertium.org/wiki/Bilingual_dictionary ; Workflow reference: https://wiki.apertium.org/wiki/Workflow_reference . Jurafsky and Martin, Speech and Language Processing, Machine Translation: https://web.stanford.edu/~jurafsky/slp3/13.pdf . Apertium is cited as a richer dictionary-using pipeline, not as a pure substitution system. The house example and Hazel Labels case are original teaching material.
Previous: Dictionary Next: Dictionary coder