Does morphology study sentence structure?

Its central focus is word structure, while syntax studies how words combine. The two interact, particularly in agreement and grammatical feature interpretation.

Is a lemma always made by deleting a suffix?

No. Irregular relationships and spelling changes can require lexical knowledge or other analysis.

Can morphology improve AI automatically?

It can provide useful structure, but benefit depends on the model, language, task, data, and implementation. Evaluate the actual system.

AI Glossary / Natural language and speech / Beginner

Morphology

Definition

Morphology is the study of word structure and the systematic relationships between word forms. In natural language processing, morphological analysis identifies information such as a word's base, its internal components, and grammatical features expressed by its form.

What it means

The words “repair,” “repairs,” and “repaired” share a lexical base but do different grammatical work. “Repairable” is related too, yet creates a different lexical expression. Morphology examines these relationships rather than treating every spelling as wholly unrelated.

This glossary entry concerns linguistic morphology. In image processing, mathematical morphology refers to a different family of operations on shapes and sets; sharing the word does not make the methods equivalent.

Word structure is not always a row of visible suffixes. Languages can use prefixes, internal changes, repetition, tone, and other patterns. Some forms also have more than one analysis until their sentence context is known. A system must interpret the form in its language rather than assume English patterns apply everywhere.

How it works

An analyzer may use a lexicon and rules, a statistical model, or a neural model to propose analyses. It can map a surface word to a lemma and grammatical features, or segment a word into meaningful components. A generator works in the other direction, producing a form from a base and specified features.

Inflection concerns grammatical variants, such as plural or past-tense forms. Derivation forms related lexical words, such as an adjective from a verb. The boundary has complexities, and not every derivation changes word class.

Annotation systems can represent features independently of visible segments. Universal Dependencies, for example, provides categories such as Number, Tense, Person, and Case, with language-specific application. Its “universal” inventory does not mean every feature occurs in every language or that a single tagger handles them all equally well.

Three practical examples

Search: connect “inspection” with relevant inflected forms without automatically erasing distinctions between inspecting and being inspected.
Translation: preserve grammatical number or tense when the target language requires a different word form.
Writing assistance: check agreement in a sentence while retaining the user's intended person, number, and time reference.

Think of it this way

Think of morphology as examining a word's design and the family of forms it belongs to. Looking only at the printed label misses how a grammatical change alters its role. The analogy is limited: natural language contains irregular patterns and ambiguity, so word families do not behave like perfectly standardized manufactured parts.

Worked example

Compare “The repairs were expensive” with “Lea repairs bicycles.” Both sentences contain the identical surface word “repairs.” In the first, it is a plural noun; in the second, it is a present-tense verb used with a third-person singular subject.

A contextual analyzer might represent the first as lemma=repair, part of speech=NOUN, Number=Plural. The second can be lemma=repair, part of speech=VERB, Tense=Present, Person=3, Number=Singular. The precise feature set depends on the annotation scheme.

A rule that simply removes final “s” obtains “repair” in both cases but loses the contrasting analyses. Conversely, preserving only the original spelling cannot expose them. The useful output pairs the original form with its contextual analysis. This example has two word occurrences and two grammatical interpretations; it does not require two different written suffixes.

Illustrative case study

Fictional case: Brook Manuals uses an assistant to generate maintenance instructions in several languages. The team initially checks only whether the correct equipment names appear. Reviewers find that some generated instructions use the wrong agreement forms, even though the vocabulary is recognizable.

The evaluation is expanded to include grammatical features in context and examples reviewed by proficient speakers. Original sentences remain available for comparison. This illustrates why vocabulary coverage alone is not a complete language-quality measure; no quantified improvement is claimed.

Common misconceptions and mistakes

Morphology is not identical to stemming. A stemmer may remove endings heuristically, whereas morphological analysis can distinguish grammatical features and contextual alternatives. Lemmatization selects a dictionary-style base, but does not by itself provide a complete analysis.

Do not equate a model's subword tokens with morphemes, or assume every final letter sequence is an affix. Irregular forms need more than blanket suffix removal. Also avoid treating an ambiguous form as evidence that one reading is always correct: “repairs” must be interpreted in its sentence.

Business use

Choose the depth of analysis according to the task. Broad search may need normalization, while translation and instruction generation need distinctions that normalization can remove. Keep the original surface form, lemma, and feature information in separate fields when they serve different purposes.

Test irregular forms, unfamiliar vocabulary, and context-dependent analyses in each supported language. Review failures involving negation, tense, number, and agreement because they can change the practical meaning of an answer. A system that produces fluent-looking words still needs evaluation on the exact distinctions the workflow relies on.

Check your understanding

A system maps both noun “repairs” and verb “repairs” to “repair.” Has it necessarily preserved all relevant information?

Answer: No. The base form alone omits the difference between a plural noun and a present-tense verb in the example. Whether that loss matters depends on the application.

Related terms

Morpheme
Lemma
Stemming
Morphological typology
Part-of-speech tagging
inflection
derivation
grammatical agreement

Learn next

take an ambiguous word form and place it in two sentences with different grammatical roles. Write the original form, lemma, and features separately. Then compare a stemmer, a lemmatizer, and a morphological analyzer to see which distinctions each output actually preserves.

Lemma

Sources

Sources: SIL, Morphology, linguistic scope: https://glossary.sil.org/term/morphology . Universal Dependencies, Universal features, feature inventory and language-specific scope: https://universaldependencies.org/u/feat/index.html . Jurafsky and Martin, Speech and Language Processing, Chapter 2, word structure: https://web.stanford.edu/~jurafsky/slp3/2.pdf . The repairs comparison and Brook Manuals case are original.

Previous: Morphological typology Next: Mosaic effect