Estonian NLP Grammar

Finite state and Constraint Grammar based analysers, proofing tools and other resources

View the project on GitHub giellalt/lang-est-x-utee

Morphology

Multicharacter symbols

Sõnaliik / Part of speech

Genitiivatribuut pole eraldi kategooria / No special tag for genitive attribute : angoora+N+Sg+Gen

Alamkategooriad / Sub-pos categories

Võrdlusastmed / Comparation

Sõnamuutus / Inflection

Käändsõna / Noun

Arv / number

Kääne / Case

Verb

Verbil on finiitsed (pöördelised) ja infiniitsed (nn käänduvad) vormid. Pöördelistel vormidel on kategooriad: tegumood, aeg, kõneviis, isik, arv, kõneliik (tegumood ja aeg on ka mõnedel infiniitsetel vormidel).

Verbs have finite (conjugable) and infinite (“declinable”) forms. The finite forms have categories: voice, tense, mood, person, number, negation (some infinite forms also have voice and tense).

Pöördelised vormid / Conjugable forms

Tegumood / Voice

Aeg / Tense

Lihtminevik = minevik / Past imperfect = past

Täisminevik / Past perfect: olema (pres) + nud/tud/dud (olen teinud)

Enneminevik / Past pluperfect: olema (impf) + nud/tud/dud (olin teinud)

Kõneviis / Mood

Pööre ja arv / Person and number

Kõneliik / Polarity

Infiniitsed (nn käänduvad) verbivormid / Infinite (“declinable”) verb forms

Kategooriate võimalikud kombinatsioonid / Possible combinations of categories

allikas / source : Heiki-Jaan Kaalep. Eesti verbi vormistik. Keel ja kirjandus 1/2015 lk 1-15.

The categories are given in the order in which the allomorphs (if they can be distinguished) that represent them are attached to the word stem (note that the treatment of allomorphs is sloppy here). The justification is that the categories are not equal, but form an hierarchy: those closer to the word end tend to be more optional, more often non-specified.

  1. voice: personal vs. impersonal (0-morph vs. t/d(aiu)), eg. elaks vs. elataks, elav vs. elatav
  2. tense: present vs. past (0-morph vs. s/si/nu), e.g. elan vs. elasin, elaks vs. elanuks
  3. mood: indicative vs. conditional vs imperative vs quotative (0-morph vs. ks vs. k/g(ue) vs. vat)
  4. person+number: notice that in personal present and past conditional, imperative and quotative moods one form is underspecified for person+number
  5. negation: affirmative vs. negative. It manifests itself via lexical means: it is either present in an exceptional wordform (some forms of olema) or gets adhered to a form, normally used in affirmative, from an immediately preceding word ei, ega or ära (e.g. ei elaks). The only case when negation has a dedicated form of its own is impersonal present indicative negative (e.g. ei elata).

The forms of negative words pole and ära are included in order to capture really all the combinations; also note that for olema, there are some constellations of categories that are the same for a wordform beginning with ole- and pole-

—- Personal finite forms —-

\

NB! ei ole = pole

\

\

\

\

\

\

NB! ei olnud = polnud

\


elanud: elama+V+Pers+Prt+Imprt: 1.11.2016 ei genereerita ega tunta ära / not recognized nor generated as of Nov 1, 2016

\

—- Personal infinite forms —-

(on, oli, …) + V+Pers+Prt+Prc = some analytical personal form

—- Impersonal finite forms —-

NB! ei olda = polda

\

\

\

NB! ei oldud = poldud

—- Impersonal infinite forms —-

(on, oli, …) + V+Impers+Prt+Prc = some analytical personal form

—- Infinite forms with no voice category —-

Exceptional cases:

personal present (Prs not implemented????), 3 words: kuulukse, tunnukse, näikse

Negation verb

Analytical forms (olen elanud, olin elanud, oleksin elanud, ei olnud elanud, ei olnuks elanud etc) are not treated here…

Partikkel -gi/-ki / Particle -gi/-ki

Kasutust (sagedust jm) iseloomustavad märgendid / Usage-related tags

Paralleelvormide erinevat kasutussagedust iseloomustavad / Usage info for parallel forms (either correct according to the norm, or incorrect)

Oletatav analüüs / Guessed analysis

Tuletus / Derivation

Eesliited on harvad, aga ikkagi … / Prefixes are seldom used, but still…

Järelliited / Suffixes

V –> N

V –> A

N –> A

A –> Adv

A –> A

N –> N

N Prop –> N

A –> N, V –> N

N –> Adv

N –> N, A –> A

Num –> N

Muud / Other tags

Copied from Sami root.lexc

Flag diacritics

Use the following flag diacritics to control downcasing of derived proper nouns (e.g. Finnish Pariisi -> pariisilainen). See e.g. North Sámi for how to use these flags. There exists a ready-made regex that will do the actual down-casing given the proper use of these flags. | @U.Cap.Obl@ | Allowing downcasing of derived names: deatnulasj. | @U.Cap.Opt@ | Allowing downcasing of derived names: deatnulasj.

These are used for limiting the number of components in a compound word (the compound transducer is cyclic, but in reality there is a limit to the length of words) | @D.Part@ | No part of a compound should have been encountered yet | @P.Part.One@ | Indicate that this could be the first part of a compound | @R.Part.One@ | Require that the first part has been encountered; if a lemma has it, it means that the lemma cannot be part2 or part3 of a compound | @D.Part.One@ | Require that the first part has NOT been encountered | @P.Part.Two@ | Indicate that this could be the second part of a compound | @R.Part.Two@ | Require that the second part has been encountered | @P.Part.Three@ | Indicate that this could be the third part of a compound | @P.Part.Bad@ | Indicate that this cannot be a part of a compound | @R.Part.Bad@ | Require that the first part has been encountered; if a lemma has it, it means that the lemma cannot be part2

POS is used for filtering derivations and compounds | @R.POS.NumCardCompound@ | compound numeral (viis+sada - five hundred)

Case is used for filtering derivations and compounds

Remember there has been some derivation from A or N; for filtering compounds derivation from V is called paradigmatic and does not result in Der; just a new POS…

Remember the stem type; for filtering compounds

A special condition that is used for filtering derivations and compounds

Tokeniser

Guesser

Symbols that need to be escaped on the lower side (towards twolc):

Sami GT convention

Legitimate strings that are not words: numbers, acronyms, …

Lexicons

For modelling compounds, the simplex word fst is concatenated with itself. For this, Kleene star operation is used, i.e. fst is concatenated zero to any number of times. For the lookup process, this creates a possibility of infinitely many passes through the fst, thus allowing infinitely long words. For limiting and controlling the passes, flag diacritics are used. Lookup process remembers which paths it has taken, and counts the passes. For remembering, it sets up flags on the path:

  1. Flag the first 3 passes through the simplex fst (thus forbidding any more passes) (the lexicon contains conventionalised compounds, thus the max number of compound components is larger than three; however, no Estonian word with more than five components has been found yet)
  2. In every pass, flag the POS and inflection, to be used in the following pass Different paths may correspond to the same surface string, e.g. mootoriõlilik = (mootori+õli)+lik and (mootori)+(õli+lik)

Lexicon-based passes

strictly simplex word; cannot be a part of a compound a simplex word, or the first part of a compound

Guesser assumes that there is only one pass, and that only the final part is important (out-of-vocabulary simplex words are treated elsewhere)

— end guesser

lexicon-based

strictly simplex words; cannot be a part of a compound

This (part of) documentation was generated from src/fst/morphology/root.lexc