Skip to content
EANC / LINGUISTICS

Linguistics

Lexical and morphological annotation of Standard Eastern Armenian.

EANC texts are supplied with lexical and morphological information: lemmas, parts of speech, inflection types and grammatical features. The following sections describe the grammatical dictionary, the classification of Eastern Armenian inflection, the annotation process and the interpretation of ambiguous analyses.

Grammatical dictionary

The grammatical dictionary is the lexical foundation of EANC. Each record identifies a lemma and its part of speech, with an inflectional class, lexical properties, accepted spelling variants and English meanings where available. Distinct meanings or parts of speech can have separate records even when their spelling is identical.

An excerpt from the grammatical dictionary
#LexemeTransliterationPOSLexical featuresInflection typeEnglish meanings
1միmiPART—Non-inflectingdon't; now!
2միmiNUMcardNon-inflectingone; some
3բանbanNinanimN11affair; thing; pursuit
4գրելgrelVtrV11write; write (for)
5հայրhayrNapl+,anim,humN71father
6սերserNinanimN92love
7սերserNinanimN11cream
8սեղանseġanNinanimN12table
The excerpt shows selected published dictionary records. N = noun; V = verb; NUM = numeral; PART = particle. Lexical features include card (cardinal), tr (transitive), inanim (inanimate), anim (animate), hum (human) and apl+ (the lexical associative-plural policy marker).

Sources of the grammatical wordlist

The original wordlist combined several complementary sources. Its description identified the following dictionaries and lexical collections:

  1. E. Galstian’s Armenian–Russian dictionary (1985) and part of E. Aghaian’s dictionary of Modern Armenian (1976), together contributing over 70,000 entries.
  2. H. Grgearian and N. Harutyunian’s dictionary of geographic names (1987–1989), contributing about 4,000 entries.
  3. A collection of common first names and family names, with about 1,000 entries.
  4. D. Gyurdjinian and N. Hekekian’s dictionary of Armenian acronyms (2007), contributing about 2,000 entries.

These figures describe the original wordlist, rather than current dictionary totals. Lexemes found in corpus texts, including neologisms absent from these sources, supplemented the inventory. Combining sources reduced gaps in morphological recognition.

Assigning parts of speech and inflection types turned the wordlist into a grammatical dictionary. Lexical properties describe the entry: for example, a personal name, an abbreviation or an animate noun. The morphological model supplies features of the particular wordform, such as plural or ablative. Current dictionary editing also records accepted spelling variants, sources and individual corrections.

English meanings and glosses

An English translation can give several meanings or explanations. A lexical gloss selects a representative equivalent. Morphological glossing separately describes the grammatical pieces of a form. These English aids support readers and language learners, but they do not translate complete sentences; missing English does not make a morphological analysis invalid.

Explore a word and its corpus occurrences to connect its lexical description with attested usage.

Morphological model

The model describes how Eastern Armenian wordforms are built from lexical entries, stems and inflectional patterns. A class such as N11 selects a pattern; shared rules supply number, case and definite or possessive endings. The model also handles licensed orthographic alternatives, irregular stems and productive formations involving numerals and abbreviations.

Characteristic nominal forms
#TypeLexemesg, nomsg, obl (gen/dat)sg, inspl, nom
1N11բանբանբան+իբան+ովբան+եր
2N12սեղանսեղանսեղան+իսեղան+ովսեղան+ներ
3N71հայրհայրհորհոր+ովհայր+եր
The + sign marks the joining of a stem and an ending. N11 and N12 contrast the plurals -եր and -ներ. N71 illustrates the stem alternation հայր / հոր. These characteristic forms follow the original inflection reference and are also licensed by the current model.
A characteristic verbal paradigm
#TypeLexemeInfinitivePerfective converbAorist, 3sgImperative, 2sg
1V11գրելգր+ելգր+ելգր+եցգր+իր
In this conjugation, the infinitive and perfective converb have the same written form. Their interpretation depends on context.

Browse inflection types on a separate reference page. Class labels connect dictionary entries, morphological analyses and grammatical searches.

Grammatical categories and token markup

EANC uses compact tags for parts of speech, nominalization, case, number, determination, comparison, converbs, participles, valency-changing categories, tense, aspect, mood, polarity and person. Inflectional class is also recorded. The notation follows the Leipzig Glossing Rules where applicable, supplemented by Armenian-specific categories.

Selected tags, Armenian labels and examples
#Tag or combinationDescriptionTraditional Armenian labelExample
1NNounԳոյականսեղան
2AAdjectiveԱծականգեղեցիկ
3VVerbԲայկարդալ
4PRONPronounԴերանունես
5ablAblativeԲացառականքաղաքից
6insInstrumentalԳործիականքաղաքով
7plPluralՀոգնակիքաղաքներ
8nmlzNominalized attributeԳոյականացված ածականգեղեցիկը
9cvb, ipfvImperfective converbԱնկատար դերբայ Iկարդում
10ptcp, resResultative participleՀարակատար դերբայկարդացած
The Armenian labels and examples carry forward the original tag reference. Codes here use current notation. A combination such as cvb, ipfv describes an imperfective converb.

A tag describes a supplied analysis, rather than independently proving how a word functions in its sentence. Search labels can combine several tags. For example, the Prospective participle requires both ptcp and fut; the full interpretation cannot be read from a single label alone.

Armenian-specific analytical decisions

  1. Genitive and dative. In nouns, the traditional genitive/dative distinction depends substantially on the ability to attach a definite article: the dative commonly has one, while the genitive does not. The original parser therefore assigned the paired Gen/Dat label to syncretic forms. Current annotation uses the shared oblique label where appropriate. Personal pronouns distinguish genitive and dative forms; personal genitives and possessive entries now receive separate treatment.
  2. Converbs. EANC uses this term broadly for dependent verbal forms without person–number agreement, including forms used with auxiliaries as well as in subordinate predications.
  3. Destinative and prospective. The traditional destinative in -ելու/-ալու overlaps formally with genitive/dative infinitives. The prospective participle in -ելիք/-ալիք is distinct. The legacy destinative query notation should not be treated as a description of every current selector.
  4. Associative plural. Forms such as Վարդանանք and Շուշանենք refer to a named person and their associates. The legacy discussion also includes pronominal group forms such as մերոնք.
  5. Nominalization and relational forms. Headless adjectives, participles and genitives can acquire nominal inflection. The legacy relational examples սեղանինը, սեղանինինը and սեղանինով illustrate layered forms meaning roughly “the one on the table”, “to that one” and “by that one”. Their analysis involves nominalization as well as case and determination.
Relational forms in the original grammatical description
#Armenian formLegacy segmentationMeaning
1սեղանինըseġan-i-n-əthe one that is on the table
2սեղանինինըseġan-i-n-in-əto the one that is on the table
3սեղանինովseġan-i-n-ovby the one that is on the table
These examples show why a sequence of endings requires a linguistic analysis. In the original discussion, the first article-like element functions as a nominalizer; further nominal marking can follow it.

Transliteration

Armenian script is the default. Corpus search display settings offer EANC transliteration and Hübschmann–Meillet. The reference table also includes basic Eastern Armenian IPA. EANC transliteration uses forms such as ǰ for ջ, ġ for ղ and ṙ for ռ.

Transliteration represents written characters: Երևան begins with E in EANC transliteration, regardless of its initial pronunciation. It does not infer unwritten vowels or contextual sound changes. The IPA option is also a character-level approximation. Changing display mode leaves queries and stored Armenian text unchanged.

View the transliteration reference
Armenian characters and current display equivalents
#ArmenianEANCHübschmann–MeilletBasic IPA
1աaaɑ
2բbbb
3գggɡ
4դddd
5եeee
6զzzz
7էēêe
8ըəǝə
9թtʽtʽtʰ
10ժžžʒ
11իiii
12լlll
13խxxχ
14ծcct͡s
15կkkk
16հhhh
17ձjjd͡z
18ղġłʁ
19ճččt͡ʃ
20մmmm
21յyyj
22նnnn
23շššʃ
24ոooo
25չčʽčʽt͡ʃʰ
26պppp
27ջǰǰd͡ʒ
28ռṙr̄r
29սsss
30վvvv
31տttt
32րrrɾ
33ցcʽcʽt͡sʰ
34ւwwv
35ուuuu
36ևewewev
37փpʽpʽpʰ
38քkʽkʽkʰ
39օōōo
40ֆfff

Annotation

Annotation connects grammatical knowledge with individual occurrences. EANC retains the original distinction between document descriptions, lexical and morphological token markup, and punctuation and sentence boundaries. Bibliographic descriptions and subcorpus selection are explained under Corpus composition.

From text to analysis

  1. Tokenization. Text is normalized and divided into sentences, words and punctuation, retaining alignment with the normalized text. Armenian inline punctuation, abbreviations, compounds and suffixes require specific boundary decisions; morphology helps resolve them.
  2. Morphological parsing. The dictionary and paradigms supply all complete licensed readings of a wordform. Each reading can carry its lemma, part of speech, grammatical features, inflectional class, segmented form, gloss and lexical English meanings.
  3. Contextual disambiguation. Constraint Grammar (CG) applies sentence-level rules to remove incompatible readings. Unresolved alternatives survive. CG cannot supply an analysis missing from morphology.

The original EANC Parser, developed by Corpus Technologies, analyzed wordforms independently of context. The current processing layer adds contextual CG. Its output remains morphological annotation rather than a syntactic parse.

Parsing example

The original worked example comes from Raffi’s Samvel. It remains useful because a single sentence illustrates ordinary inflection, verbal constructions and lexical ambiguity.

Այդ կողմից երբեմն փայլատակում էր կայծակը, և լսելի էր լինում որոտման խուլ դղրդյուն, որ գուշակում էր հորդ անձրև։

The table retains the lexical readings discussed in the legacy example. It describes historical parser output, not a fresh analysis by the current dictionary and CG. Commas and the final Armenian full stop are punctuation tokens.

Read the complete legacy word-by-word example
Raffi example: legacy lexical and morphological readings
#WordformLemma or alternativesAnalysis
1ԱյդայդDemonstrative pronoun: that
2կողմիցկողմNoun, singular ablative: side
3երբեմներբեմնAdverb: sometimes
4փայլատակումփայլատակել / փայլատակումImperfective converb / noun: glittering
5էրէAuxiliary, past, singular, 3rd person
6կայծակըկայծակNoun, singular nominative, definite: lightning
7ևևConjunction: and
8լսելիլսելիAdjective: audible
9էրէAuxiliary, past, singular, 3rd person
10լինումլինելImperfective converb: be
11որոտմանորոտումNoun, singular genitive/dative
12խուլխուլAdjective: muffled / noun: tumour
13դղրդյունդղրդյունNoun, singular nominative: rumble
14որորInterrogative/relative pronoun: which
15գուշակումգուշակ / գուշակել / գուշակումNoun, locative / imperfective converb / noun, nominative
16էրէAuxiliary, past, singular, 3rd person
17հորդհայր / հոր / հորդFather, dative with 2nd-person possessive / pit, nominative with 2nd-person possessive / adjective: abundant
18անձրևանձրևNoun, singular nominative: rain

The legacy XML represented sentences with <se>, words with <w> and punctuation with <c>. Within each word, <n> recorded the number of analyses; alternative <gloss> blocks carried lemma, lexical, grammatical, inflectional and translation fields. Current storage differs, but the distinction between the written occurrence and its alternative analyses remains important.

CG: selecting readings in context

In Աղմուկը դադարել էր։, morphology supplies infinitive and perfective-converb readings of դադարել. The following auxiliary supports the perfective converb, allowing CG to remove the infinitive competitor. Other rules address negative auxiliaries, prohibition, modal constructions, postpositions, attributive contexts and numeral combinations.

Աղմուկը դադարել էր։

The noise had stopped.

Contextual selection for դադարել before the auxiliary էր
#Morphological candidateCG result in this sentence
1V, inf · infinitiveRemoved: the auxiliary supports the perfective construction.
2V, cvb, pfv · perfective converbRetained
The table contrasts the bare infinitive and converb candidates. Nominally marked infinitive candidates are governed by a separate annotation policy.

A surviving reading remains a candidate. CG makes grammatical searches more selective, but does not resolve every lexical ambiguity or identify every construction.

Periphrastic verbal constructions

Auxiliaries and dependent verb forms remain separate tokens. To find a construction such as an imperfective present, combine an imperfective-converb condition with a nearby present-tense auxiliary, rather than asking a single token to carry the entire construction. Inspect the context to distinguish the intended construction from coincidental co-occurrence.

Coverage and ambiguity

Recognized forms and unknown words

Recognition means that the model supplies an analysis; it does not establish that the analysis is correct in context. An unknown word has no supplied reading. Recognition varies across collections, orthographies and registers.

The original parser discussion identified recent loans, neologisms, Russian or English code-switching, abbreviations, proper names, technical terms, Western Armenian or obsolete spellings, inflectional variants, OCR errors and original typos as causes of zero analysis. These remain useful diagnostic categories, although their historical frequencies do not measure the current corpus.

The legacy parser sometimes proposed a hypothetical analysis for an abbreviation followed by a hyphen and inflectional ending, marking its lemma with a question mark. Current productive recognition is licensed by explicit morphological rules; that historical question-mark convention should not be assumed for current output.

Ambiguity and search interpretation

Ambiguity can be regular, as with the infinitive and perfective converb of գրել, or lexical, as with հարգի, which can be an adjective meaning “respectable” or a verbal form of հարգել “respect”. A grammatical search can match a supplied reading that is not the intended reading of the occurrence.

Contextual constraints can reduce this noise. Restricting a query to unambiguous occurrences can also exclude genuine examples: ambiguity is a property of the annotation, not evidence that the intended construction is absent. The original parser also suppressed some highly improbable competitors, including a locative asum from as; this is a historical policy example rather than a claim about the current rule inventory.

Historical coverage and ambiguity: January 2009

These figures are preserved from the legacy parser account. Percentages refer to its historical token populations, including special-token categories. Reported rounding and the 99% Other written total are retained. They are not current performance figures.

Legacy parser distribution, January 2009 — percentage of tokens in each column
#CategoryFictionSciencePressOther writtenOralEANC total
11 analysis75.4%67.0%72.6%69.4%65.2%72.6%
22 analyses15.1%9.6%12.7%12.3%12.8%13.3%
33 analyses1.8%2.1%2.0%1.8%1.6%1.9%
44–7 analyses1.8%2.1%2.0%1.8%1.6%1.9%
5All ambiguous18.7%13.7%16.7%15.8%16.0%17.0%
6Hypothetical0.0%1.3%0.6%0.7%0.2%0.5%
7Not recognized5.4%11.9%7.7%7.0%12.3%7.5%
8Cyrillic, Latin, digits0.3%6.3%2.8%5.6%6.0%2.4%
9Reported total100%100%100%99%100%100%

The original annotation overview separately reported lexical and morphological markup for over 90% of tokens and English translations for about 85%. Those statements describe the legacy release and use different measures from the distribution above.

Evaluation and limits

Coverage, correctness and disambiguation are different measurements. CG evaluation asks both how many occurrences a rule affects and whether it removes valid readings. The current rule set has changes that still require renewed corpus evaluation; earlier assessments are not a blanket accuracy guarantee.

Recorded CG judgments are AI contextual assessments with retained evidence, rather than an expert-reviewed gold standard. Checks that search results reproduce stored annotations establish search fidelity; they do not independently establish linguistic correctness. Interpret an occurrence alongside its full sentence and source.

Dictionary, morphology and CG revisions can change future analyses. Published corpus generations retain their own linguistic inputs, so descriptions of the latest resources need not describe every currently indexed occurrence.