UD for Hebrew
Tokenization and Word Segmentation
- In general, words are delimited by whitespace characters. Description of the exceptions follows.
- According to typographical rules, many punctuation marks are attached to the preceding space-delimited word. We always represent them as separate tokens (words); this holds even for hyphenated compounds such as בין-משרדית ‘inter-office’ (three tokens).
- Abbreviations marked with double quotation mark, such as צה”ל ‘IDF’, are not tokenized, and are considered a single token.
- Some function words in Hebrew (commonly known as משהוכלב) are attached as prefixes to the following token, but we represent them as separate tokens:
- Oblique case markers: ב, ל, כ, מ ( ‘of’, ‘as’, ‘to’ and ‘in’ respectively).
- The conjunction marker ו - ‘and’
- The definite determiner ה - ‘the’
- The subordination marker ש - ‘that’
- Combinations of the above. Note that not all elements are overt. The definite marker _ה, when appearing after the case markers ב or ל, is covert e.g. בבית may be analyzed as both ‘at home’ -> ב_+ה_+בית ‘in+the+home’ and ‘at a house’ -> ב_+בית ‘at+house’. As a result, there is inherent ambiguity in space-delimited tokens in Hebrew where tokens starting with _ב or ל can carry two different analyses, with or without the definite marker.
- We also separate pronominal clitics to their lemmas in the following cases
- Prepositions and case markers: שלנו ‘our’ -> של_ + _אנחנו ‘of we’
- Verbs inflected to mark their accusative object אהבתיה ‘I loved her’ -> אהבתי_ + _את_ + _היא loved+acc+her’.
- Genitive Nouns ספרו ‘his book’ -> ספר_ + _של_ + _הוא ‘book+of+he’. To indicate in the FORM column that these case markers and pronominal in fact undergo morphological fusion in the raw text, we distinguish them from the standard pronominals and case markers using underscore at the direction of fused elements.
Morphology
Tags
- Hebrew uses all universal POS categories, but particles (PART).
- In Hebrew, DET represents definite articles (ה ‘the’), quantifiers (הרבה ‘many’), distributive determiners (כל ‘every’) and interrogatives ( אילו ‘which’). However, demonstratives (זה ‘this’) are tagged as PRON. We separate possessive determiners into a possessive marker של ‘of’ which is ADP and the relevant pronoun PRON.
- Hebrew auxiliaries (AUX) are modal and aspectual expressions that combine with the bare infinitive of the main verb, such as חייב ‘must’ and יכול ‘can’. Note that modals do not constitute any uniform syntactic class in Hebrew, and there is an ongoing debate as to the POS of each modal expression. The various inflections of the היה ‘be’ verb with respect to gender, person, negation and tense (only past and future) are also tagged as AUX.
- Existential markers, either present tense יש\אין or past/future inflections of the היה ‘be’ verb, are tagged as VERB.
- Participles in Hebrew are formed as present tense verbs and they are used as either nouns (NOUN), or modals (AUX). Note that deverbal expressions (whether nouns or adjectives) are tagged as (VERB) when they are complemented, but as (NOUN) or (ADJ) when they are not.
Features
- Abbr: The Abbreviation feature (Abbr=Yes) may be marked for NOUN (ח”כ ‘Knesset member’), PROPN (ארה”ב ‘United States’), ADJ (ז”ל ‘R.I.P’) and ADP (ע”י ‘by’).
- Case: has possible values of Acc, Gen and Tem. Gen and Acc mark accusative and genitive prepositions (ADP) respectively. When a pronoun is segmented into its basic lemmas (e.g. אותנו -> את_ + _אנחנו) the Case=Acc feature accompanies the case marker את. Temporal case (Tem) in Hebrew appears in temporal subordination כש ‘when (+clause)’, whose POS tag is SCONJ.
- Definite: The Definiteness feature has two values: Def and Cons, for definite and for construct states respectively. The feature Definite=cons marks nouns (NOUN), determiners (DET), verbs (VERB), numerals (NUM) and adjectives (ADJ) that act as the first token in a construct-state construction (מבנה סמיכות), and they mark the resulting construct-state NP as inherently definite (i.e., not requiring an overt ה definite marker). Definite=Def also marks instances of pronouns which express similarity (e.g. אנשים במרכז אותה תנועה).
- Gender: All nominal words (NOUN, PROPN and PRON) in Hebrew have a grammatical Gender: either feminine (Fem) or masculine (Masc). The following parts of speech also inflect for Gender because they must agree with nouns: ADJ, VERB, AUX and NUM. Nouns and verbs which are underspecified for gender are marked Gender=Fem,Masc.
- HebBinyan: Hebrew has a morphological system of root+binyan conjugation that conveys agency (simple, intensive or causative) and voice (active passive, or middle/reflexive) in verbs (VERB).
- HebExistential: marked as
Yes
if given to existential verbs (VERB) יש ‘there is’ or אין ‘there is no’ or past/future inflections such as היה ‘there was’ תהיה ‘there will be’. - Mood: the value Imp marks imperative verb forms.
- Number: The values for Number can be: sing (singular), plur (plural), dual for a couple (שנתיים ‘two years’, as opposed to שנים ‘years’), or dual,plur for items which only have a dual form and no plural form (hence it is not necessarily two - could be more, e.g. שוליים ‘margins’). The following parts of speech also inflect for number (either singular or plural) because they must agree with nouns: ADJ, DET, VERB, AUX, where dual and plural inflect the same way.
- Person: Finite verbs (VERB or AUX) and pronouns (PRON) are marked for 1, 2 or 3 person in the case of first, second and third person respectively. Verbs are also marked , or Person=1,2,3 when on the case of lemmas unspecified for person (for example in present tense).
- Polarity: neg or pos are given to copular verbs according to their polarity (negative/positive).
- Prefix: Yes is given to non-standalone adverbials (ADV) which are either loan prefixes (e.g. מולטי ‘multi’, אנטי ‘anti’ etc.) or Hebrew adverbials with the same function (e.g. בלתי, אי ‘non’).
- PronType: relevant for pronouns (PRON) and determiners (DET). PronType can take the following values: Prs for personal pronouns (e.g. אתה ‘you’, היא ‘she’ etc), Dem for demonstrative pronouns ( זה ‘it’, ‘this’), Int for interrogative pronouns (אילו ‘which ones’ מיהו ‘who’), and Ind for indefinite pronouns (e.g איכשהו, מישהו somehow, someone). When describing determiners, PronType can be either Art for articles (in hebrew there is a definite article ה_ ‘the’), or Int for interrogative determiners (מי, כיצד, האם, ‘who’, ‘how’ ‘whether’ respectively). Note that in Hebrew האם ‘whether’ can also appear in the matrix clause.
- Reflex: PRON can take a Reflex=Yes feature that indicates reflexivity (e.g. עצמי ‘myself’).
- Tense: marks a VERB when it is either Past or Fut (future). Present tense is not explicitly marked with this particular feature.
- VerbForm: marks a verb if it is either in infinitive form (inf) or participle (part).
- VerbType=Mod is given to modal auxiliaries (AUX) and VerbType=Cop is given to copular auxiliaries.
- Voice: can be either for active (act), passive (pass) or unaccusative (mid) verb forms.
- HebSource: This feature can mark any category and accepts the values ConvUncertainHead or ConvUncertainLabel to indicate that the head or label information for this token, respectively, is based on unreliable information detected in the automatic conversion process. In UD v2.8, this attribute is moved from FEATS to MISC.
Syntax
Word Order
The canonical word order in Hebrew is SVO. However, this generalization is flexible, as accusative and oblique case marking allow free order with respect to subject, verb and objects. Furthermore, there are various specific constructions in which different word orders are more common (מחר יגיע אורח ‘tomorrow will arrive a guest’ is more in use than מחר אורח יגיע ‘tomorrow a guest will arrive’ though both are perfectly grammatical). In other constructions the SVO word order is ungrammatical, such as the Hebrew existential construction which surfaces as VOS (יש לו המיומנות הנחוצה להיות מושל).
Subjects and other Arguments
Subjects are labeled nsubj and have the following characteristics:
- Case marking: Subjects occur in nominative case without adpositions.
- A clause is labeled csubj when it serves as the subject of its matrix clause. Participles are tagged nsubj when they serve as subjects of the clause: המגדל נחש בביתו, אל יופתע אם ימצא עצמו מוכש Whoever raises a snake at home should not be surprised if stung.
Arguments have the following characteristics:
- Case marking: Arguments defined in the Hebrew grammar may be bare noun phrases in accusative, or prepositional phrases in accusative, dative, oblique or genitive. In Hebrew, arguments additional to direct object are always prepositional phrases. Hence, direct objects are labeled obj, and prepositional phrases are marked obl, whether they are core arguments or adjuncts.
- Passivization: Objects become subjects when verbs are passivized. In accordance with v2, they are labeled nsubj.
- If a verb subcategorizes for the infinitive (e.g. phrasal verbs or verbs of control), the infinitival complement is labeled xcomp.
Non-verbal Clauses
- The copular verb להיות ‘be’ is used in attributional, locative and possessive nonverbal clauses when either non-finite or inflected to past/future. Hebrew allows for demonstrative determiners to serve as copulas as well, e.g. שוקולד זה טעים ‘chocolate is tasty’.
- In equational clauses the copular verb may appear also with inflection to present tense הילד הזה הוא אני ‘this boy is me.’
- Existential clauses use a different lemma, יש ‘give’ with an accusative object: יש אוכל ‘there is food.’
- In addition to the verbal copula, Hebrew should have 4 additional pronominal ones:
- היה [haya] – the proper, verbal copula, used in past and future contexts
- הוא [hu] – the 3rd person masculine personal pronoun ‘he’, present tense use, used just as in Arabic, agrees with predicate (but just the one masculine singular lemma is needed)
- זה [ze] – masculine demonstrative ‘this’, present tense use, does not need to agree, probably borrowed from Polish to, Russian это / èto, & co.
- הינו [hino] – declinable particle, present tense, formal (historically probably from a locative stem, synchronically only used as a copula or rarely by itself for strong existential, in which case it would not be AUX)
- אינו [eino] – lexically negative equivalent of the previous item (by itself can also mean someone has died, i.e. ‘is no more’, but again, this would not be tagged cop)
Relations Overview
- The following relation subtypes are used in Hebrew:
- acl:relcl for the head of a relative clause
- case:acc for an accusative case marker
- case:gen for a genitive case marker
- compound:affix non-standalone yet orthographically separated lemmas, largely originated as loan affixes (אנטי וירוס ‘antivirus’) but not only (e.g. בלתי הפיך ‘irreversible’).
- compound:smixut for a noun modifying another noun in a construct state.
- flat:name for the second part of a complex name
- mark:q for yes-no interrogative clause markers (either matrix or subordinate).
- nmod:poss for a covert pronoun in a genitive construction
- nsubj:cop for the subject of a copular sentence. The copular verb in Hebrew is optional. Consequently, the only way to retrieve copular sentences is by their only obligatory element, which is the subject.
- The following main types are not used alone and must be subtyped: compound, flat
- The following relation types are not used in Hebrew at all: clf, expl.
Treebanks
Currently, Hebrew has 3 treebanks: