home edit page issue tracker

UD for Swiss German

Introduction

This is a copy of the current German documentation UD for German, which we generally follow for Swiss German. This introduction explains the most important differences which influence the annotation.

Please check the readme/GitHub repo of the GSW treebank for further/current information.

Differences to German UD Guidelines

As for German, words are generally delimited by white spaces. However, there is a lot more freedom in merging any words together, which can’t usually be split in an easy way. I.e. we use the German tokenization and introduce a separate tag for merged words (see meta tag TAG+ described further down).

The POS annotations are generally based on the German guidelines, namely the Stuttgart-Tübingen-TagSet (STTS) and some changes according to the TIGER annotation scheme. Furthermore, dealing with Swiss German, there is the need for an additional POS tag PTKINF, not present in the STTS tagset, as well as for the “meta tag” TAG+.

• PTKINF is an infinitive particle which does not exist in Standard German but is frequently used in dialects. It comes in the form of go, cho, goge, lo to name a few, as in Si gönd go poschte. (They go shopping.) In the Standard German translation, Sie gehen einkaufen., we can see that there is no equivalent. Concerning dependencies it is treated as a marker introducing a finite clause subordinate to another clause (mark) because they usually appear in um … zu (in order to) constructions in German.

• TAG+ is used to handle merged words; we introduced the “+“-sign which can be added to any PoS tag. In the STTS there is one tag like this: the APPRART, used for combinations of articles and prepositions like im consisting of in + dem (in the). However, in Swiss German these kind of merges are performed with any kind of words and just merging the tags would result in a big tagset. Therefore we decided to use the “head” of the word or the first word as tag and simply add a plus to show that this word incorporates another one Hollenstein and Aepli, 2014. Like this, they can easily be found and, if needed, manually expanded. Frequent examples of such words include hemmer (haben + wir), häts (hat + es), and sinz (sind + sie), for we have, it has and they are.

The Universal Dependency POS (UPOS) tags are converted according to the mapping provided by the Universal Dependency. Additionaly:

• PTKINF are converted to PART
• the plus sign in TAG+ are disgarded

UD for German

Tokenization and Word Segmentation

• In general, words are delimited by whitespace characters. Description of exceptions follows.
• According to typographical rules, many punctuation marks are attached to a neighboring word. We usually tokenize them as separate tokens (words) with the exception of abbreviations such as usw. “etc.” which are kept as one token with the period.
• German compounds are written as one word and we do not split them.
• There is one class of multi-word tokens: the contractions of prepositions and definite articles. Example: zum = zu + dem “to the”.

Morphology

Tags

• German uses all 17 universal POS categories, including particles (PART).
• The following words are particles in German: nicht “not”, and the infinitive marker zu “to”.
• The pronoun (PRON) vs. determiner (DET) distinction is based on word lists because the traditional grammar does not define determiners. In general, words that inflect for gender, to be able to agree with a modified noun, are tagged DET, even if they act independently in a given sentence; that includes possessives.
• German auxiliary verbs (AUX) are:
  • sein for perfect tenses of some verbs (ich bin gekommen “I have come”) and as copula (er ist alt “he is old”)
  • haben for perfect tenses of the remaining verbs (ich habe gegessen “I have eaten”)
  • werden for future tense (ich werde reisen “I will travel”) and for the passive (es wurde gegessen “it was eaten”)
  • modal verbs dürfen “may”, können “can”, mögen “want”, müssen “must”, sollen “shall”, wollen “want”
  • The verbs sein, haben and werden can also occur as normal verbs (VERB), meaning “be, have, become”.
• There are four main (de)verbal forms, distinguished by the UPOS tag and the value of the VerbForm feature:
  • Infinitive Inf, tagged VERB or AUX.
  • Finite verb Fin, tagged VERB or AUX.
  • Participle Part, tagged VERB or AUX.
  • Verbal noun Vnoun, tagged NOUN, looks like a capitalized infinitive but has an article and may inflect.

Nominal Features

• Nominal words (NOUN, PROPN and PRON) have an inherent Gender feature with one of three values: Masc, Fem or Neut.
  • The following parts of speech inflect for Gender because they must agree with nouns: ADJ, DET, VERB, AUX. For verbs (including auxiliaries), only participles can inflect for gender. Finite verbs don’t.
• The two main values of the Number feature are Sing and Plur. The following parts of speech inflect for number: NOUN, PROPN, PRON, ADJ, DET, VERB, AUX (finite and participles).
• Case has 4 possible values: Nom, Gen, Dat, Acc. It occurs with the nominal words, i.e., NOUN, PROPN, PRON, ADJ, DET. However, case forms of nouns are extremely ambiguous and most of the time the case is distinguished only by the form of the article.
• Definite has 2 values: Ind, Def. It is used to distinguish the indefinite and definite articles (DET).

Degree and Polarity

• Degree applies to adjectives (ADJ) and adverbs (ADV) and has one of three possible values: Pos, Cmp, Sup.
• Polarity is used to mark the negative particle nicht, i.e., only the Neg value is used.

Verbal Features

• Finite verbs always have one of three values of Mood: Ind, Imp or Sub (called Konjunktiv in German).
• Indicative and subjunctive verbs always have one of two values of Tense: Past, Pres.
  • In the subjunctive mood, the tense feature is used to distinguish Konjunktiv I (Pres) and Konjunktiv II (Past).
  • Imperative forms do not have the Tense feature.
  • The Tense feature is also used to distinguish present and past participles (kommend “coming” vs. gekommen “come”).
• The features Aspect and Voice are not used in German because both the perfect aspect and the passive voice are expressed periphrastically.

Pronouns, Determiners, Quantifiers

• PronType is used with pronouns (PRON), determiners (DET) and adverbs (ADV).
• NumType is used with numerals (NUM), adjectives (ADJ) and determiners (DET).
• The Poss feature marks possessive personal determiners (e.g. mein “my”), possessive interrogative or relative determiners (e.g. wessen “whose”).
• The Reflex feature is always used together with PronType=Prs and it marks reflexive pronouns (mich/mir, dich/dir, sich, uns, euch). Note that their forms in the first and second person are ambiguous with irreflexive accusative forms, and the Reflex feature must be decided by context.
• Person is a lexical feature of personal pronouns (PRON) and has three values, 1, 2 and 3. With personal possessive determiners (DET), the feature actually encodes the person of the possessor. Person is not marked on other types of pronouns and on nouns, although they can be almost always interpreted as the 3rd person.
  • As a cross-reference to subject, person is also marked on finite verbs (VERB, AUX).
• The Polite feature distinguishes informal second-person pronouns (du, ihr, Polite=Infm) from the formal Sie (Polite=Form). The formal pronoun is phonologically equivalent in all its case forms to the third-person plural sie but it is distinguished in orthography by the capital letter S. We tag it as second person (because that is its meaning) and we do not tag its number (because it is used both for singular and plural addressees) despite the fact that it combines with third-person plural verbs. The parser must learn that Person=2|Polite=Form subject attaches to Number=Plur|Person=3 verbs, while Number=Sing|Person=2|Polite=Infm subject attaches to Number=Sing|Person=2 verbs.
• There are two layered features, Gender[psor] and Number[psor]. They appear with certain possessive determiners and encode the lexical gender/number of the possessor. The extra layer is needed to distinguish these lexical features from the inflectional gender and number that mark agreement with the modified (possessed) noun.

Other Features

• Besides the layered features listed above, there are several other language-specific features:
  • Typo
• The following universal features are not used in German: Animacy, Evident.

Syntax

Core Arguments, Oblique Arguments and Adjuncts

• Nominal subject (nsubj) is a noun phrase in the nominative case, without preposition.
  • A finite subordinate clause may serve as the subject and is labeled csubj.
  • If an infinitive verb is to serve as the subject, it becomes a verbal noun (it gets the neuter singular nominative article and it is capitalized in writing), thus it is labeled nsubj.
• Objects defined in the German grammar may be bare noun phrases in accusative, dative or genitive, or prepositional phrases in accusative, dative or genitive. For the purpose of UD the objects are divided to core objects, labeled obj or iobj, and oblique objects, labeled obl:arg.
  • Bare accusative, dative and genitive objects are considered core.
  • All prepositional objects are considered oblique.
  • Accusative objects of some verbs alternate with finite clausal complements, which are labeled ccomp.
  • If a verb subcategorizes for the infinitive (e.g. phasal verbs or verbs of control), the infinitival complement is labeled xcomp.
  • If a verb subcategorizes for two core objects, one of them accusative (or ccomp) and the other non-accusative, then the non-accusative object is labeled iobj. Core nominal objects in other situations are labeled just obj.
• Adjuncts (or, following the German grammar, adverbial modifiers realized as noun phrases) are usually prepositional phrases, but they can be bare noun phrases as well. They are labeled obl:
  • Temporal modifiers realized as accusative noun phrases: ich arbeite jeden Tag “I work every day.”
  • All prepositional phrases that are not prepositional objects (i.e., their role and form is not defined lexically by the predicate) are adjuncts.
• Extra attention has to be paid to the reflexive pronoun sich. It can function as:
  • Core object (obj or iobj): er sah sich im Spiegel “he sighted himself in the mirror.”
  • Reciprocal core objects (obj or iobj): sie küssten sich “they were kissing each other.”
  • Inherently reflexive verb, cannot exist without the reflexive clitic, and the clitic cannot be substituted by an irreflexive pronoun or a noun phrase. In accord with the current UD guidelines, we label the relation between the verb and the clitic as expl:pv, not compound. Example: wir müssen uns beeilen “we must hurry.”
• In passive clauses, the subject is labeled with nsubj:pass or csubj:pass, respectively.
  • The auxiliary verb in periphrastic passive is labeled aux:pass.
  • If the demoted agent is present, it has the form of a bare instrumental phrase and its relation is labeled obl:agent.

Non-verbal Clauses

• The copula verb sein (be) is used in equational, attributional, locative, possessive and benefactory nonverbal clauses.
  • Existential clauses use a different verb, geben (give) with an accusative object: es gibt Essen “there is food.”

Relations Overview

• The following relation subtypes are used in German:
  • nsubj:pass for nominal subjects of passive verbs
  • csubj:pass for clausal subjects of passive verbs
  • obl:agent for agents of passive verbs
  • obl:arg for prepositional objects
  • expl:pv for reflexive clitics of inherently reflexive verbs
  • aux:pass for passive auxiliaries
  • compound:prt for separable verb prefixes
  • det:poss for possessive determiners
  • nmod:poss for possessive modifier phrases
• The following relation types are not used in German at all: clf, dislocated

Treebanks (Swiss German)

There is one Swiss German UD treebank:

• Swiss German-UZH

© 2014–2024 Universal Dependencies contributors. Site powered by Annodoc and brat

.