Are prepositions always about physical location?

No. They can express temporal and abstract relationships as well as spatial ones.

Does ADP mean only preposition?

No. The Universal Dependencies category also includes postpositions and other adpositional constructions.

Can identifying the preposition resolve an ambiguous sentence?

Not by itself. The system may still need to determine what the full phrase modifies and which relationship is intended.

AI Glossary / Natural language and speech / Beginner

Prepositions

Definition

Prepositions are adpositions that come before the phrases they accompany. They help express relationships such as location, direction, time, and association. In “under the table,” under precedes the noun phrase “the table.” The grammatical relationship matters as well as position: a word is not a preposition merely because it appears before a noun.

What it means

Prepositions are often short, but they can make a large difference to meaning. Compare an item being “in the box,” “beside the box,” and “under the box.” The object named by the noun phrase stays the same while its relationship to the item changes.

Adposition is the broader category that also includes postpositions, which follow their associated phrases. Languages differ in their preferred constructions. A translation may therefore express an English prepositional relationship through a postposition, case morphology, or another structure rather than a matching word in the same position.

How it works

An analysis identifies the preposition, the phrase it accompanies, and the larger expression to which that construction relates. In “Mina worked in the workshop,” the location phrase contributes information about the working event. In “the cabinet in the workshop,” a similar phrase helps identify the cabinet.

Attachment can be ambiguous. A parser must decide which relationship a prepositional phrase contributes, using the sentence and available context. Recognizing the word's category alone does not settle the meaning of the whole phrase.

Annotation conventions also matter. Universal Dependencies uses ADP for adpositions, while language-specific guidelines resolve words with several uses and certain borderline constructions. For example, a word appearing before a noun phrase can behave differently when it introduces a clause. A project should follow its documented scheme rather than assuming that a familiar spelling always receives one universal analysis. Multiword expressions need explicit handling because their combined relationship may not be represented adequately by labeling one isolated word.

Three practical examples

1. Document extraction: A system distinguishes “delivery to the depot” from “delivery from the depot” so that destination and source are not confused.

2. Search interpretation: A query assistant checks whether a location phrase describes the requested object or the event being discussed.

3. Translation quality: A reviewer verifies that a translated relationship preserves the intended meaning without forcing English word order onto the target language.

Think of it this way

Think of a preposition as a small directional label connecting a phrase to something else in the sentence. Changing the label can change the relationship. The analogy is limited because many uses are abstract or idiomatic, and knowing a label's common spatial meaning does not predict every construction in which it appears.

Worked example

Consider the original sentence “Mina photographed the technician with a drone.” Ignoring punctuation, its seven tokens are numbered as follows: Mina(1), photographed(2), the(3), technician(4), with(5), a(6), drone(7).

The preposition with is at position 5 and its following noun phrase, “a drone,” occupies positions 6-7. The full prepositional phrase spans positions 5-7. One reading makes the drone an instrument used in photographing. Another associates the drone with the technician being photographed.

Both readings keep the same words and phrase span. What differs is the phrase's attachment and interpretation. Suppose a toy evaluation includes two separate contexts, one establishing each intended reading. A rule that always attaches this phrase to the photographing event would get one of those two context-specific decisions right: 1/2=50%. That invented score illustrates the limitation of the stated rule; it is not a benchmark result or an estimate of how often either reading occurs naturally.

Illustrative case study

Fictional case: Orchard Inventory extracts movements from short warehouse notes. Reviewers find that its early prototype records the same relationship for “from the loading area” and “to the loading area.” It has recognized the place name but discarded the preposition's contribution.

The team builds paired examples that keep the location constant while changing the relation. It also tests sentences with more than one location phrase, checking each attachment separately. This fictional case illustrates why a small word can determine a useful data field; it does not claim a real deployment or measured improvement.

Common misconceptions and mistakes

Do not treat prepositions as disposable words whenever the task depends on relationships. Removing them can erase distinctions between source and destination or between different spatial arrangements. Whether a text-processing pipeline can omit them depends on what it must preserve.

Also avoid assuming every prepositional phrase has one obvious attachment, or that each preposition has one fixed translation. A correct word label is only part of a correct sentence analysis. Meaning may depend on the associated phrase, the governing expression, and the broader context.

Business use

For business language systems, test prepositions through meaningful contrasts. Keep the entities constant while changing the relationship, and check whether the extracted result changes appropriately. Include ambiguous sentences with clarifying context so evaluation measures interpretation rather than rewarding a fixed attachment preference that happens to fit one example.

Check your understanding

In the drone example, do the two readings require different words or different phrase spans?

Answer: No. The same prepositional phrase can receive different attachments and interpretations. Context is needed to determine which reading is intended in a particular use.

Related terms

Postposition
Attachment ambiguity
Noun phrase
Semantic role
Part-of-speech tagging
adposition
prepositional phrase
dependency parsing

Learn next

Learn noun phrases and prepositional phrases, then compare pairs such as source versus destination. Next study attachment ambiguity using complete sentences with context. Finally, consult a language's annotation guidelines to understand how broad grammatical concepts map onto its tags and relations.

Attachment ambiguity

Sources

Universal Dependencies, ADP definition and category clarifications: https://universaldependencies.org/u/pos/ADP.html

Matthew S. Dryer, WALS chapter 85, Order of Adposition and Noun Phrase: https://wals.info/chapter/85

Bird, Klein, and Loper, Natural Language Processing with Python, chapter 8, primary textbook discussion of grammar and attachment ambiguity: https://www.nltk.org/book/ch08.html

The warehouse case, drone sentence, and toy evaluation are original illustrations; their readings are explanatory possibilities rather than observed corpus frequencies.

Previous: Prepared statement Next: Preprocessing