Grammar¶

This is the core reference page of the Esperanto section. It collects the rules a rule-based pipeline needs — for the tokenizer, the lemmatizer, the POS tagger and stop-word handling. Because Esperanto's morphology has no exceptions, each rule below can be turned almost directly into code.

The guiding idea: an Esperanto word is built from a root plus grammatical endings and optional affixes. Strip the endings and affixes and you have the lemma; read the endings and you have the part of speech and the inflection.

What this page leaves out

The participles (-ant-, -int-, -ont- active; -at-, -it-, -ot- passive), the compound tenses built from them with esti, and the passive voice are just as regular, but they are not covered here. A pipeline that implements only the rules below will mis-analyse forms like leganta, legita or estas legata.

Reference

The rules and examples on this page follow the classic A Complete Grammar of Esperanto by Ivy Kellerman Reed (1910), a public-domain work included in the repository at resources/esperanto/books/esperanto_grammar.txt. See Resources § Books for the full citation.

Alphabet¶

Esperanto uses a 28-letter Latin alphabet. It is strictly phonetic: every letter is always pronounced the same way, and every sound is written with exactly one letter. Six letters carry diacritics: the five circumflexed consonants are used by no other language, and ŭ also appears in the Belarusian Latin alphabet.

Letter	Name	Sound (approx.)
`ĉ`	ĉo	ch in "church"
`ĝ`	ĝo	g in "gem"
`ĥ`	ĥo	ch in Scottish "loch"
`ĵ`	ĵo	s in "measure"
`ŝ`	ŝo	sh in "ship"
`ŭ`	ŭo	w in "how" (used in diphthongs)

The letters q, w, x, y are not part of the alphabet.

Encoding matters for tokenization

The diacritics are real Unicode letters (ĉ is U+0109), not c plus a combining mark — but text found in the wild may use either form. Two ASCII fallbacks also exist for keyboards that cannot type the diacritics:

the x-system: cx gx hx jx sx ux for ĉ ĝ ĥ ĵ ŝ ŭ;
the h-system (from the Fundamento): ch gh hh jh sh u.

A robust tokenizer should normalise input to a single canonical form (Unicode NFC, real diacritics) before applying any other rule. This project's data files keep the real diacritics and never the x-system (see resources/README.md).

Word-class endings¶

This is the single most useful rule for POS tagging and lemmatization: the final vowel of a full word marks its part of speech.

Ending	Word class	Example	Root	Meaning
`-o`	noun	`libro`	`libr-`	book
`-a`	adjective	`bona`	`bon-`	good
`-e`	adverb (derived)	`rapide`	`rapid-`	quickly
`-i`	verb (infinitive)	`kanti`	`kant-`	to sing

From one root you get the whole family: muziko (music), muzika (musical), muzike (musically). To lemmatize, strip the grammatical ending (and any inflection below) to recover the root, then re-attach the class vowel to get the dictionary form.

Grammatical inflections¶

Nouns and adjectives take two optional endings, always after the class vowel:

Ending	Meaning	Example
`-j`	plural	`libroj` (books), `bonaj` (good, pl.)
`-n`	accusative (direct object)	`libron`, `bonan`
`-jn`	plural and accusative	`bonajn librojn`

Two rules make these fully regular:

Adjectives agree with their noun in both number and case: "la bonaj libroj", "mi legas interesajn librojn".
The order is fixed: root → class vowel → -j → -n. So the parse of librojn is libr- + -o (noun) + -j (plural) + -n (accusative).

The accusative also marks direction of motion, but for morphological analysis the -n suffix is what matters.

Verb system¶

Verbs do not conjugate for person or number — only for tense and mood, with a single ending each. The same root takes any of these:

Ending	Tense / mood	Example (`kant-`, sing)
`-as`	present	`kantas` (sing / sings)
`-is`	past	`kantis` (sang)
`-os`	future	`kantos` (will sing)
`-us`	conditional	`kantus` (would sing)
`-u`	imperative / volitive	`kantu!` (sing!)
`-i`	infinitive	`kanti` (to sing)

mi kantas, vi kantas, ili kantas — the verb never changes for the subject. To lemmatize a verb, replace its tense/mood ending with -i (kantis → kanti).

The article¶

There is one article, the definite la ("the"). It is invariable — no plural, no gender, no case:

la libro (the book), la libroj (the books), la bona libro (the good book).

There is no indefinite article: libro means both "a book" and simply "book". la is a natural stop-word (see Stop-words).

Personal pronouns¶

The pronouns are a small, closed set — ideal to hard-code:

Pronoun	Meaning	Possessive (`+ -a`)
`mi`	I	`mia`
`vi`	you (sg. & pl.)	`via`
`li`	he	`lia`
`ŝi`	she	`ŝia`
`ĝi`	it	`ĝia`
`ni`	we	`nia`
`ili`	they	`ilia`
`oni`	one / people (impersonal)	`onia`
`si`	reflexive (himself/herself/itself/themselves)	`sia`

The accusative of a pronoun adds -n: min (me), vin, lin, ŝin, nin, ilin. The possessives are ordinary adjectives, so they also inflect: miajn librojn (my books, acc. pl.).

Correlatives¶

The correlatives are a strikingly regular 5 × 9 grid of 45 common words (this/that/which/some/every/no + thing/person/place/time/reason…). Each is a prefix (the meaning category) plus an ending (the type). Learn the grid and you get all 45 for free.

The five prefixes:

Prefix	Meaning
`ki-`	interrogative / relative (wh-, "which")
`ti-`	demonstrative ("that")
`i-`	indefinite ("some")
`ĉi-`	universal ("every", "all")
`neni-`	negative ("no", "none")

The nine endings and the full table:

Ending → type	`ki-`	`ti-`	`i-`	`ĉi-`	`neni-`
`-o` thing	kio	tio	io	ĉio	nenio
`-u` individual	kiu	tiu	iu	ĉiu	neniu
`-a` kind	kia	tia	ia	ĉia	nenia
`-es` possession	kies	ties	ies	ĉies	nenies
`-e` place	kie	tie	ie	ĉie	nenie
`-am` time	kiam	tiam	iam	ĉiam	neniam
`-al` reason	kial	tial	ial	ĉial	nenial
`-el` manner	kiel	tiel	iel	ĉiel	neniel
`-om` quantity	kiom	tiom	iom	ĉiom	neniom

The -u and -a correlatives take both -j and -n (tiujn, kiuj), like the nouns and adjectives they resemble. The -o correlatives take -n but never -j (kion, tion — there is no tioj), and the -e correlatives take -n to mark direction (tien, kien). The other five endings are invariable.

All 45 are stop-words, and this table is the only place the project stores them: resources/esperanto/vortoj.json holds roots, not forms, so a lemmatizer is expected to generate the grid from the two axes above rather than look words up. See resources/README.md § Roots, not forms.

A drill for the grid

esperanto12.net/en/tabelvortoj walks through the table cell by cell with example sentences — the fastest way to stop reading it as 45 separate words.

Affixes¶

Affixes are what make Esperanto vocabulary compositional — and they are the heart of lemmatization, because a long word is usually a root wrapped in affixes. They attach between the root and the grammatical ending.

Prefixes¶

Prefix	Meaning	Example
`mal-`	direct opposite	`bona` → `malbona` (good → bad)
`ge-`	both sexes together	`patro` → `gepatroj` (father → parents)
`ek-`	sudden start / brief action	`iri` → `ekiri` (to go → to set off)
`re-`	again / back	`veni` → `reveni` (to come → to return)
`dis-`	separation, scattering	`doni` → `disdoni` (to give → to distribute)
`mis-`	wrongly	`kompreni` → `miskompreni` (to misunderstand)
`bo-`	relation by marriage	`patro` → `bopatro` (father-in-law)
`pra-`	primordial / great- (kinship)	`avo` → `praavo` (great-grandfather)

Suffixes¶

Suffix	Meaning	Example
`-in-`	female	`patro` → `patrino` (father → mother)
`-ist-`	professional / adherent	`instrui` → `instruisto` (teacher)
`-ej-`	place for	`lerni` → `lernejo` (to learn → school)
`-il-`	tool / instrument	`tranĉi` → `tranĉilo` (to cut → knife)
`-ar-`	collection / set	`arbo` → `arbaro` (tree → forest)
`-et-`	diminutive (smaller)	`domo` → `dometo` (house → cottage)
`-eg-`	augmentative (bigger)	`domo` → `domego` (house → mansion)
`-ul-`	person characterised by	`juna` → `junulo` (young → youth)
`-an-`	member / inhabitant	`urbo` → `urbano` (city → citizen)
`-ec-`	abstract quality	`bona` → `boneco` (good → goodness)
`-ig-`	to make / cause	`granda` → `grandigi` (big → to enlarge)
`-iĝ-`	to become	`ruĝa` → `ruĝiĝi` (red → to blush)
`-ind-`	worthy of	`ami` → `aminda` (to love → lovable)
`-em-`	inclined to	`labori` → `laborema` (to work → hardworking)
`-aĉ-`	pejorative (bad quality)	`domo` → `domaĉo` (house → hovel)

Affixes stack. mal-san-ul-ej-o = mal- (opposite) + san- (health) + -ul- (person) + -ej- (place) + -o (noun) = hospital (literally "place for unhealthy persons"). A lemmatizer that knows this affix table can decompose such words by rule.

Numbers and compounding¶

The cardinal numbers are built from a handful of roots:

0	1	2	3	4	5	6	7	8	9	10	100	1000
`nul`	`unu`	`du`	`tri`	`kvar`	`kvin`	`ses`	`sep`	`ok`	`naŭ`	`dek`	`cent`	`mil`

Larger numbers are compounded by juxtaposition: dek du (12), dudek (20), dudek unu (21), cent tridek kvin (135). Adding endings derives related words:

ordinals with -a: unua (first), dua (second);
multiples with -obl-: duobla (double);
fractions with -on-: duono (a half), kvarono (a quarter);
collectives with -op-: duope (in twos).

More generally, any two roots can compound into one word, with the last root as the head: vapor-ŝipo (steamship), dorm-o-ĉambro (bedroom, with a connecting -o- for pronounceability). The compound's word class comes from its final ending, exactly as for a single root.

Worked examples¶

Putting the rules together, here is a sentence parsed word by word:

La juna instruistino legas interesan libron. "The young (female) teacher is reading an interesting book."

Word	Decomposition	Analysis
`La`	`la`	article (stop-word)
`juna`	`jun-` + `-a`	adjective — "young"
`instruistino`	`instru-` + `-ist-` + `-in-` + `-o`	noun — "female teacher"
`legas`	`leg-` + `-as`	verb, present — "reads" (lemma `legi`)
`interesan`	`interes-` + `-a` + `-n`	adjective, accusative — "interesting"
`libron`	`libr-` + `-o` + `-n`	noun, accusative — "book" (lemma `libro`)

A second example shows plural, accusative agreement and a correlative:

Ĉiuj miaj amikoj legas tiujn librojn. "All my friends are reading those books."

Ĉiuj (correlative ĉiu + -j), miaj (possessive mia + -j), amikoj (amik-o-j) — the subject is plural, so nothing takes -n; tiujn librojn (tiu-j-n, libr-o-j-n) is the plural accusative object, and the demonstrative agrees with its noun. This regular agreement is precisely what a rule-based tagger can verify and exploit.

Where to go next¶

Word lists — the stop-word list in full, generated from the data the library reads.
Python Library § API — how these rules become a Doc / Token interface.