Multiword expression translation begins with recognising that spaces are not meaning boundaries. Phrasal verbs, compound nouns, technical terms, light-verb constructions, institutional phrases and collocations often behave as units even though several words are visible. Translate each component independently and the target can be fluent yet semantically wrong.
People searching for compound noun translation, phrasal verb translation, technical multiword terms, collocations, fixed expressions or light-verb constructions are solving a segmentation problem before a vocabulary problem. “Take into account” is not three unrelated lexical decisions. “Machine learning model evaluation” is not four isolated nouns. Precision begins by deciding which words must move together.
This article owns the phrase-segmentation precision layer. It explains compositionality, fixed and semi-fixed expressions, phrasal verbs, light verbs, noun compounds, technical terms, collocations, institutional labels, discontinuous expressions and phrase-level QA. The central rule is: segment first, interpret second, translate third.
Meaning Units vs Word Boundaries
The first task is to identify the structural relation before selecting target vocabulary. Orthographic spaces do not reliably identify translation units. That distinction matters because translation often changes order, morphology, prepositions, case, punctuation or phrase length. The target may need to be longer or shorter than the source, but it must preserve which elements belong together, which item modifies which head, what counts as one conceptual unit, and what relation connects the parts.
The common failure is assigning a separate target equivalent to every visible source word. Such errors often survive ordinary proofreading because every target word can be individually plausible. The breakdown becomes visible only when the reviewer reconstructs the phrase, list, definition or category system as a whole. In technical, legal, educational, scientific and web content, that hidden structural shift can produce the wrong component, wrong rule, wrong category or wrong user action.
A practical method is to box words that function together before dictionary lookup For example, “take place” often means occur, not physically take + place. Before release, see whether the target translates the whole function. This creates a reusable diagnostic instead of relying on intuition. Where the source itself is ambiguous, preserve or flag that ambiguity rather than choosing the most convenient parse. Precision comes from making the same relationship available to the target reader, not from imitating the source’s visible arrangement.
Compositionality
A precise translator treats this as a mapping problem rather than a sequence of isolated words. Some phrases are predictable from parts while others are partly or wholly lexicalised. That distinction matters because translation often changes order, morphology, prepositions, case, punctuation or phrase length. The target may need to be longer or shorter than the source, but it must preserve which elements belong together, which item modifies which head, what counts as one conceptual unit, and what relation connects the parts.
The common failure is assuming transparent-looking phrases are free combinations. Such errors often survive ordinary proofreading because every target word can be individually plausible. The breakdown becomes visible only when the reviewer reconstructs the phrase, list, definition or category system as a whole. In technical, legal, educational, scientific and web content, that hidden structural shift can produce the wrong component, wrong rule, wrong category or wrong user action.
A practical method is to ask whether changing one component freely preserves normal usage For example, “make a decision” is transparent but conventionally fixed. Before release, consult dictionary/corpus entries for the phrase. This creates a reusable diagnostic instead of relying on intuition. Where the source itself is ambiguous, preserve or flag that ambiguity rather than choosing the most convenient parse. Precision comes from making the same relationship available to the target reader, not from imitating the source’s visible arrangement.
Fixed Expressions
The visible source order is only one possible surface form for the underlying relation. Fixed phrases resist component substitution and often have functional target equivalents. That distinction matters because translation often changes order, morphology, prepositions, case, punctuation or phrase length. The target may need to be longer or shorter than the source, but it must preserve which elements belong together, which item modifies which head, what counts as one conceptual unit, and what relation connects the parts.
The common failure is translating their internal words instead of their discourse function. Such errors often survive ordinary proofreading because every target word can be individually plausible. The breakdown becomes visible only when the reviewer reconstructs the phrase, list, definition or category system as a whole. In technical, legal, educational, scientific and web content, that hidden structural shift can produce the wrong component, wrong rule, wrong category or wrong user action.
A practical method is to search the whole expression first For example, “in accordance with” may map to one target preposition or legal formula. Before release, compare established target usage. This creates a reusable diagnostic instead of relying on intuition. Where the source itself is ambiguous, preserve or flag that ambiguity rather than choosing the most convenient parse. Precision comes from making the same relationship available to the target reader, not from imitating the source’s visible arrangement.
Semi-Fixed Frames
This feature becomes risky when the target language reorganises phrase or category structure. Constructions such as not only X but also Y contain stable frame plus variable slots. That distinction matters because translation often changes order, morphology, prepositions, case, punctuation or phrase length. The target may need to be longer or shorter than the source, but it must preserve which elements belong together, which item modifies which head, what counts as one conceptual unit, and what relation connects the parts.
The common failure is translating frame pieces separately and losing long-distance relation. Such errors often survive ordinary proofreading because every target word can be individually plausible. The breakdown becomes visible only when the reviewer reconstructs the phrase, list, definition or category system as a whole. In technical, legal, educational, scientific and web content, that hidden structural shift can produce the wrong component, wrong rule, wrong category or wrong user action.
A practical method is to mark fixed frame and replaceable slots For example, “the more X, the more Y” is a construction. Before release, check both halves in target. This creates a reusable diagnostic instead of relying on intuition. Where the source itself is ambiguous, preserve or flag that ambiguity rather than choosing the most convenient parse. Precision comes from making the same relationship available to the target reader, not from imitating the source’s visible arrangement.
Phrasal Verbs
A fluent target is not enough unless the same conceptual dependency remains recoverable. Verb+particle combinations can form meanings not predictable from each component. That distinction matters because translation often changes order, morphology, prepositions, case, punctuation or phrase length. The target may need to be longer or shorter than the source, but it must preserve which elements belong together, which item modifies which head, what counts as one conceptual unit, and what relation connects the parts.
The common failure is translating particles literally. Such errors often survive ordinary proofreading because every target word can be individually plausible. The breakdown becomes visible only when the reviewer reconstructs the phrase, list, definition or category system as a whole. In technical, legal, educational, scientific and web content, that hidden structural shift can produce the wrong component, wrong rule, wrong category or wrong user action.
A practical method is to test the verb-particle unit in a dictionary or paraphrase with one verb For example, “look into the issue” means investigate. Before release, replace target with a simple paraphrase and compare. This creates a reusable diagnostic instead of relying on intuition. Where the source itself is ambiguous, preserve or flag that ambiguity rather than choosing the most convenient parse. Precision comes from making the same relationship available to the target reader, not from imitating the source’s visible arrangement.
Separable Phrasal Verbs
The first task is to identify the structural relation before selecting target vocabulary. Objects can interrupt one lexical unit. That distinction matters because translation often changes order, morphology, prepositions, case, punctuation or phrase length. The target may need to be longer or shorter than the source, but it must preserve which elements belong together, which item modifies which head, what counts as one conceptual unit, and what relation connects the parts.
The common failure is failing to recognise a discontinuous phrasal verb. Such errors often survive ordinary proofreading because every target word can be individually plausible. The breakdown becomes visible only when the reviewer reconstructs the phrase, list, definition or category system as a whole. In technical, legal, educational, scientific and web content, that hidden structural shift can produce the wrong component, wrong rule, wrong category or wrong user action.
A practical method is to recombine verb and particle during source analysis For example, “turn the machine off” contains turn off. Before release, check target predicate as one unit. This creates a reusable diagnostic instead of relying on intuition. Where the source itself is ambiguous, preserve or flag that ambiguity rather than choosing the most convenient parse. Precision comes from making the same relationship available to the target reader, not from imitating the source’s visible arrangement.
Light-Verb Constructions
A precise translator treats this as a mapping problem rather than a sequence of isolated words. Verbs such as make, take, give and have can carry little lexical content while the noun names the event. That distinction matters because translation often changes order, morphology, prepositions, case, punctuation or phrase length. The target may need to be longer or shorter than the source, but it must preserve which elements belong together, which item modifies which head, what counts as one conceptual unit, and what relation connects the parts.
The common failure is using a literal target verb that sounds unnatural or changes meaning. Such errors often survive ordinary proofreading because every target word can be individually plausible. The breakdown becomes visible only when the reviewer reconstructs the phrase, list, definition or category system as a whole. In technical, legal, educational, scientific and web content, that hidden structural shift can produce the wrong component, wrong rule, wrong category or wrong user action.
A practical method is to identify the event noun and test a simple-verb paraphrase For example, “make a decision” ≈ decide. Before release, match target register and event. This creates a reusable diagnostic instead of relying on intuition. Where the source itself is ambiguous, preserve or flag that ambiguity rather than choosing the most convenient parse. Precision comes from making the same relationship available to the target reader, not from imitating the source’s visible arrangement.
Academic Support-Verb Constructions
The visible source order is only one possible surface form for the underlying relation. Formal prose uses conduct an analysis, perform an assessment and reach a conclusion. That distinction matters because translation often changes order, morphology, prepositions, case, punctuation or phrase length. The target may need to be longer or shorter than the source, but it must preserve which elements belong together, which item modifies which head, what counts as one conceptual unit, and what relation connects the parts.
The common failure is over-simplifying formal register or translating the support verb physically. Such errors often survive ordinary proofreading because every target word can be individually plausible. The breakdown becomes visible only when the reviewer reconstructs the phrase, list, definition or category system as a whole. In technical, legal, educational, scientific and web content, that hidden structural shift can produce the wrong component, wrong rule, wrong category or wrong user action.
A practical method is to compare target academic conventions For example, some languages prefer a simple verb, others a formal noun phrase. Before release, check comparable professional texts. This creates a reusable diagnostic instead of relying on intuition. Where the source itself is ambiguous, preserve or flag that ambiguity rather than choosing the most convenient parse. Precision comes from making the same relationship available to the target reader, not from imitating the source’s visible arrangement.
Compound Nouns
This feature becomes risky when the target language reorganises phrase or category structure. English noun stacks encode hidden relations with the head often at the end. That distinction matters because translation often changes order, morphology, prepositions, case, punctuation or phrase length. The target may need to be longer or shorter than the source, but it must preserve which elements belong together, which item modifies which head, what counts as one conceptual unit, and what relation connects the parts.
The common failure is reordering components without unpacking relations. Such errors often survive ordinary proofreading because every target word can be individually plausible. The breakdown becomes visible only when the reviewer reconstructs the phrase, list, definition or category system as a whole. In technical, legal, educational, scientific and web content, that hidden structural shift can produce the wrong component, wrong rule, wrong category or wrong user action.
A practical method is to expand with of/for/by clauses For example, “translation quality assurance process” = process for assuring translation quality. Before release, draw relation hierarchy. This creates a reusable diagnostic instead of relying on intuition. Where the source itself is ambiguous, preserve or flag that ambiguity rather than choosing the most convenient parse. Precision comes from making the same relationship available to the target reader, not from imitating the source’s visible arrangement.
Technical Multiword Terms
A fluent target is not enough unless the same conceptual dependency remains recoverable. Domain terms can be compositional yet still have one established target equivalent. That distinction matters because translation often changes order, morphology, prepositions, case, punctuation or phrase length. The target may need to be longer or shorter than the source, but it must preserve which elements belong together, which item modifies which head, what counts as one conceptual unit, and what relation connects the parts.
The common failure is inventing a plausible paraphrase instead of using canonical terminology. Such errors often survive ordinary proofreading because every target word can be individually plausible. The breakdown becomes visible only when the reviewer reconstructs the phrase, list, definition or category system as a whole. In technical, legal, educational, scientific and web content, that hidden structural shift can produce the wrong component, wrong rule, wrong category or wrong user action.
A practical method is to research authoritative target-domain sources For example, “working memory” is a specific psychological construct. Before release, record the approved term in the termbase. This creates a reusable diagnostic instead of relying on intuition. Where the source itself is ambiguous, preserve or flag that ambiguity rather than choosing the most convenient parse. Precision comes from making the same relationship available to the target reader, not from imitating the source’s visible arrangement.
Institutional Phrases
The first task is to identify the structural relation before selecting target vocabulary. Organizations use stable expressions that function as labels. That distinction matters because translation often changes order, morphology, prepositions, case, punctuation or phrase length. The target may need to be longer or shorter than the source, but it must preserve which elements belong together, which item modifies which head, what counts as one conceptual unit, and what relation connects the parts.
The common failure is using a near-synonym that names a different policy or service. Such errors often survive ordinary proofreading because every target word can be individually plausible. The breakdown becomes visible only when the reviewer reconstructs the phrase, list, definition or category system as a whole. In technical, legal, educational, scientific and web content, that hidden structural shift can produce the wrong component, wrong rule, wrong category or wrong user action.
A practical method is to check official target-language institutional usage For example, “quality assurance” may have a standard translation. Before release, search site/document family for consistency. This creates a reusable diagnostic instead of relying on intuition. Where the source itself is ambiguous, preserve or flag that ambiguity rather than choosing the most convenient parse. Precision comes from making the same relationship available to the target reader, not from imitating the source’s visible arrangement.
Collocations
A precise translator treats this as a mapping problem rather than a sequence of isolated words. Words form conventional partnerships even when meaning remains transparent. That distinction matters because translation often changes order, morphology, prepositions, case, punctuation or phrase length. The target may need to be longer or shorter than the source, but it must preserve which elements belong together, which item modifies which head, what counts as one conceptual unit, and what relation connects the parts.
The common failure is combining dictionary equivalents into unnatural or misleading target phrases. Such errors often survive ordinary proofreading because every target word can be individually plausible. The breakdown becomes visible only when the reviewer reconstructs the phrase, list, definition or category system as a whole. In technical, legal, educational, scientific and web content, that hidden structural shift can produce the wrong component, wrong rule, wrong category or wrong user action.
A practical method is to check how target experts naturally pair the words For example, English says strong evidence, heavy rain and make a mistake. Before release, use target corpora or authoritative texts. This creates a reusable diagnostic instead of relying on intuition. Where the source itself is ambiguous, preserve or flag that ambiguity rather than choosing the most convenient parse. Precision comes from making the same relationship available to the target reader, not from imitating the source’s visible arrangement.
Verb-Preposition Frames
The visible source order is only one possible surface form for the underlying relation. Verbs select complements such as depend on, refer to and consist of. That distinction matters because translation often changes order, morphology, prepositions, case, punctuation or phrase length. The target may need to be longer or shorter than the source, but it must preserve which elements belong together, which item modifies which head, what counts as one conceptual unit, and what relation connects the parts.
The common failure is translating the preposition independently. Such errors often survive ordinary proofreading because every target word can be individually plausible. The breakdown becomes visible only when the reviewer reconstructs the phrase, list, definition or category system as a whole. In technical, legal, educational, scientific and web content, that hidden structural shift can produce the wrong component, wrong rule, wrong category or wrong user action.
A practical method is to learn the predicate with its argument frame For example, a target language may use case marking rather than a preposition. Before release, check grammar around target verb. This creates a reusable diagnostic instead of relying on intuition. Where the source itself is ambiguous, preserve or flag that ambiguity rather than choosing the most convenient parse. Precision comes from making the same relationship available to the target reader, not from imitating the source’s visible arrangement.
Adjective-Preposition Frames
This feature becomes risky when the target language reorganises phrase or category structure. Expressions such as responsible for and similar to form conventional lexical-grammatical units. That distinction matters because translation often changes order, morphology, prepositions, case, punctuation or phrase length. The target may need to be longer or shorter than the source, but it must preserve which elements belong together, which item modifies which head, what counts as one conceptual unit, and what relation connects the parts.
The common failure is choosing a target adjective but keeping the English preposition pattern. Such errors often survive ordinary proofreading because every target word can be individually plausible. The breakdown becomes visible only when the reviewer reconstructs the phrase, list, definition or category system as a whole. In technical, legal, educational, scientific and web content, that hidden structural shift can produce the wrong component, wrong rule, wrong category or wrong user action.
A practical method is to lookup target adjective with complements For example, the target may require a different case or particle. Before release, read examples from native sources. This creates a reusable diagnostic instead of relying on intuition. Where the source itself is ambiguous, preserve or flag that ambiguity rather than choosing the most convenient parse. Precision comes from making the same relationship available to the target reader, not from imitating the source’s visible arrangement.
Multiword Conjunctions
A fluent target is not enough unless the same conceptual dependency remains recoverable. As long as, because of, in spite of and in order to function as relation units. That distinction matters because translation often changes order, morphology, prepositions, case, punctuation or phrase length. The target may need to be longer or shorter than the source, but it must preserve which elements belong together, which item modifies which head, what counts as one conceptual unit, and what relation connects the parts.
The common failure is translating internal words rather than logical relation. Such errors often survive ordinary proofreading because every target word can be individually plausible. The breakdown becomes visible only when the reviewer reconstructs the phrase, list, definition or category system as a whole. In technical, legal, educational, scientific and web content, that hidden structural shift can produce the wrong component, wrong rule, wrong category or wrong user action.
A practical method is to identify relation type first For example, “in spite of” encodes concession. Before release, compare clause relation. This creates a reusable diagnostic instead of relying on intuition. Where the source itself is ambiguous, preserve or flag that ambiguity rather than choosing the most convenient parse. Precision comes from making the same relationship available to the target reader, not from imitating the source’s visible arrangement.
Named Entities
The first task is to identify the structural relation before selecting target vocabulary. Organization, product and program names can contain ordinary words while functioning as proper-name units. That distinction matters because translation often changes order, morphology, prepositions, case, punctuation or phrase length. The target may need to be longer or shorter than the source, but it must preserve which elements belong together, which item modifies which head, what counts as one conceptual unit, and what relation connects the parts.
The common failure is translating a brand or institution into a non-existent name. Such errors often survive ordinary proofreading because every target word can be individually plausible. The breakdown becomes visible only when the reviewer reconstructs the phrase, list, definition or category system as a whole. In technical, legal, educational, scientific and web content, that hidden structural shift can produce the wrong component, wrong rule, wrong category or wrong user action.
A practical method is to check official target form For example, World Health Organization has established multilingual names. Before release, use authoritative naming source. This creates a reusable diagnostic instead of relying on intuition. Where the source itself is ambiguous, preserve or flag that ambiguity rather than choosing the most convenient parse. Precision comes from making the same relationship available to the target reader, not from imitating the source’s visible arrangement.
Discontinuous Expressions
A precise translator treats this as a mapping problem rather than a sequence of isolated words. Some units are separated by inserted material. That distinction matters because translation often changes order, morphology, prepositions, case, punctuation or phrase length. The target may need to be longer or shorter than the source, but it must preserve which elements belong together, which item modifies which head, what counts as one conceptual unit, and what relation connects the parts.
The common failure is missing the relationship because pieces are distant. Such errors often survive ordinary proofreading because every target word can be individually plausible. The breakdown becomes visible only when the reviewer reconstructs the phrase, list, definition or category system as a whole. In technical, legal, educational, scientific and web content, that hidden structural shift can produce the wrong component, wrong rule, wrong category or wrong user action.
A practical method is to mark long-distance frame before drafting For example, “take the new evidence into account.” Before release, check that target expresses consider as one predicate. This creates a reusable diagnostic instead of relying on intuition. Where the source itself is ambiguous, preserve or flag that ambiguity rather than choosing the most convenient parse. Precision comes from making the same relationship available to the target reader, not from imitating the source’s visible arrangement.
Phrase Families
The visible source order is only one possible surface form for the underlying relation. Related expressions form reusable patterns around one concept. That distinction matters because translation often changes order, morphology, prepositions, case, punctuation or phrase length. The target may need to be longer or shorter than the source, but it must preserve which elements belong together, which item modifies which head, what counts as one conceptual unit, and what relation connects the parts.
The common failure is translating each occurrence independently and creating inconsistent collocation. Such errors often survive ordinary proofreading because every target word can be individually plausible. The breakdown becomes visible only when the reviewer reconstructs the phrase, list, definition or category system as a whole. In technical, legal, educational, scientific and web content, that hidden structural shift can produce the wrong component, wrong rule, wrong category or wrong user action.
A practical method is to build phrase families in a glossary For example, make/reach/reverse a decision, decision-making process. Before release, search repeated family across document. This creates a reusable diagnostic instead of relying on intuition. Where the source itself is ambiguous, preserve or flag that ambiguity rather than choosing the most convenient parse. Precision comes from making the same relationship available to the target reader, not from imitating the source’s visible arrangement.
The Replacement Test
This feature becomes risky when the target language reorganises phrase or category structure. A multiword expression often has a simple-word paraphrase that reveals its unit status. That distinction matters because translation often changes order, morphology, prepositions, case, punctuation or phrase length. The target may need to be longer or shorter than the source, but it must preserve which elements belong together, which item modifies which head, what counts as one conceptual unit, and what relation connects the parts.
The common failure is mistaking word count for semantic complexity. Such errors often survive ordinary proofreading because every target word can be individually plausible. The breakdown becomes visible only when the reviewer reconstructs the phrase, list, definition or category system as a whole. In technical, legal, educational, scientific and web content, that hidden structural shift can produce the wrong component, wrong rule, wrong category or wrong user action.
A practical method is to replace whole phrase with one-word paraphrase privately For example, “carry out an investigation” → investigate. Before release, ensure target preserves event and register. This creates a reusable diagnostic instead of relying on intuition. Where the source itself is ambiguous, preserve or flag that ambiguity rather than choosing the most convenient parse. Precision comes from making the same relationship available to the target reader, not from imitating the source’s visible arrangement.
The Reverse-Chunk Test
A fluent target is not enough unless the same conceptual dependency remains recoverable. Source and target units need not align one-to-one in word count but should align in meaning. That distinction matters because translation often changes order, morphology, prepositions, case, punctuation or phrase length. The target may need to be longer or shorter than the source, but it must preserve which elements belong together, which item modifies which head, what counts as one conceptual unit, and what relation connects the parts.
The common failure is leaving source chunks unmapped or adding target chunks without source basis. Such errors often survive ordinary proofreading because every target word can be individually plausible. The breakdown becomes visible only when the reviewer reconstructs the phrase, list, definition or category system as a whole. In technical, legal, educational, scientific and web content, that hidden structural shift can produce the wrong component, wrong rule, wrong category or wrong user action.
A practical method is to segment both source and target independently and align units For example, three source words may map to one target verb. Before release, investigate unmapped chunks. This creates a reusable diagnostic instead of relying on intuition. Where the source itself is ambiguous, preserve or flag that ambiguity rather than choosing the most convenient parse. Precision comes from making the same relationship available to the target reader, not from imitating the source’s visible arrangement.
Practical Diagnostic Workflow
Before drafting a dense passage, separate the source into meaningful units. Mark the head of each phrase, draw connections to modifiers or members, identify any fixed multiword units, and note whether a list or definition has hierarchy. Translate only after the structure is stable. Then reverse the analysis on the target: can another reader reconstruct the same head, membership, hierarchy and category boundaries without seeing the source?
This approach also helps with AI-assisted translation. Generative systems are often excellent at local fluency and can therefore conceal structural drift. They may attach a phrase to the nearest noun, split a technical compound, flatten a nested list or replace a category label with a friendlier but broader word. Use AI to propose wording, then validate the conceptual graph independently.
Recovery When Structure Has Drifted
When an error appears, return to the smallest phrase, list or definition that contains the broken relationship. Rebuild its internal structure first, then choose target wording. Avoid global replacement until you know whether the same source string has the same function everywhere. After repairing the local unit, inspect surrounding references, headings, tables and internal links because structural terminology often propagates across a document.
Quality-Assurance Checklist
- Box multiword units before lookup.
- Recognise phrasal verbs and light verbs.
- Unpack compound nouns.
- Research technical terms.
- Use official institutional names.
- Check collocations in target usage.
- Translate selected preposition frames as units.
- Mark discontinuous expressions.
- Maintain phrase-family consistency.
- Do not equate word count with information count.
- Run reverse-chunk alignment.
- Verify AI-generated technical compounds.
Frequently Asked Questions
What is a multiword expression?
A sequence of words that functions as a conventional or meaningful unit in ways that make independent word-by-word translation unreliable.
Are all idioms multiword expressions?
Many are, but the category is much broader and includes compounds, phrasal verbs, collocations, technical terms and light-verb constructions.
Should a multiword source phrase stay multiword?
No. It may become one word, another phrase or a grammatical construction in the target.
Why are English noun compounds difficult?
Because they compress relations that many target languages express overtly.
Can translation memory solve phrase segmentation?
It helps with repeated segments, but terminology and phrase-level analysis are still needed.
Can AI identify phrase boundaries?
Often, but rare compounds and technical terms can be mis-segmented. Verify against domain evidence.
Where This Fits in the eduKate Translation Architecture
Use the Context Stack, the terminology-consistency owner, the ambiguous-words owner and the modifier-attachment lane for adjacent problems. Vocabulary collocation and lexical depth route into the Vocabulary Learning Hub.
Final Principle
Spaces are not meaning boundaries. Recognise the phrase, compound, predicate frame or technical term before translating its parts. Then choose the target unit—one word or many—that carries the same concept and function.