home edit page issue tracker

This page still pertains to UD version 1.

Specific constructions

Existential sentences

In Turkish, existential sentences are (primarily) formed using var ‘there is/are’ and its negative counterpart yok. Although they do not exactly function like adjectives, following METU-Sabancı treebank, we mark them as ADJ and treat them like other nominals that require a copular suffix to function as predicates. Note that, in the most common use, the 3rd person singular present copular marker is null. In general, we indicate this wit an additional IG, which marks the “predicatehood” of the verb, and also is in parallel with the other copular constructions.

We analyze this constructions like nominal predicates. var and yok act as the head of the predicate, the thing that “exists” is the subject:

Evde bira var -0 . \n There is beer at home.
cop(var, -0)
nsubj(var, beer)
nmod(var, evde)

The same construction is used for expressing someone ‘have/has’ something or not. In this case, we mark the possessor with nmod:own (not to be confused with ``nmod:poss` used for nominal compounds with possessive marking).

Benim hiç param yok -tu . \n I did not have any money (lit: there was no money of mine)
cop(yok, -tu)
nsubj(yok, param)
advmod(yok, hiç)
nmod:own(yok, benim)

Note that when possessor is not expressed, whether the sentence is an existential (‘there is/are’) or a possession statement (‘something has’) can be understood from the possessive marking (or lack of it) on the subject. (Q: should this be indicated by the dependency relations)

In subordinate clauses the copula ol- is used in existential predicates. In this case, we mark ol- as the head of the subordinate (existential) clause.

Evde ekmek olma -dığını bilmiyordum \n I did not know that there was no bread at home
nsubj(olma, ekmek)
nmod(olma, evde)
mark(olma, -dığını)
ccomp(bilmiyordum, olma)
Ali'nin parası ol -duğu gün \n the day that Ali has/had money
nsubj(ol, parası)
nmod:own(ol, Ali'nin)
mark(ol, -düğü)
acl(gün, ol)

Besides subordinate clauses, the copula ol- can be used in a similar way (the meaning is slightly different than the cases where var or yok is used). Again, in these cases, we mark ol- as the head of the clause.

Bunun bir bedeli olur. \n This would have a consequence
nsubj(olur, bedeli)
det(bedeli, bir)
nmod:own(olur, bunun)

Noun compounds

Turkish noun compounds are generally classified into three classes:

  1. Bare compounds: tahta kapı ‘wooden door’, kadın subay ‘woman army officer’, İspanyol ressam ‘Spanish painter’, Güniz Sokak ‘Güniz street’
  2. -(s)I compounds, possessive marked head with bare complements: otobüs bileti ‘bus ticket’, Tuz Gölü ‘Salt Lake’, Asya ülkesi ‘Asian country’.
  3. genitive-posssessive construction: arabanın motoru ‘engine of the car/car’s engine’, babamın işi ‘my father’s job’, Ali’nin arabası ‘Ali’s car’, Benim arkadaşlarım ‘My friends’.

We link the first kind with compound, while the other two are linked with nmod:poss.

Note that the compounds can be complex with arbitrary structure, as in ‘[[İstanbul üniversites-i]-nin [tahta kapı]-sı]’.

Furthermore, the nominals can include subordinate clauses, like [Ali’nin henüz görmediği] film ‘the movie (that) Ali hasn’t seen yet’. In these cases, the relation is ccomp xcomp.

We need more specific guidelines, or maybe a new relation for marking compounds with subordinate clauses.

Yes/no questions an the question particle

Yes-no questions in Turkish is formed by question particle -mI (mı/mi/mu/mü). UD currently does not have a clear way of marking question particles.

We mark question particles as AUX, and (for now) introduce a new feature Question that is set for question particles as well as other words that introduce questions (PronType=Int is also used, but likely it is not sufficient/suitable in this case).

The relation between the main predicate and the question mark is aux (although we may want to subtype the relation and use aux:q in the future).

(??) If the -mI is attached to a copula, we chain the relations (the question particle is headed by the copula, not the main predicate).

Besides forming general yes/no questions by attaching to a predicate, question particle also attaches to the other constituents in a sentence. For example,

Declarative sentence

Yarın siz Ankara'ya gideceksiniz. \n You will go to Ankara tomorrow.
Yarın siz Ankara'ya gidecek misiniz. \n Will you go to Ankara tomorrow?
Siz Ankara'ya yarın mı gideceksiniz. \n Will you go to Ankara TOMORROW? 
Siz yarın Ankara'ya mı gideceksiniz. \n Will you go TO ANKARA tomorrow?
Yarın Ankara'ya siz mi gideceksiniz. \n Will YOU go to Ankara tomorrow?

Although there is a preference to place the question particle and the word modifies close to the verb, other word orders are also possible:

Yarın siz mi Ankara'ya gideceksiniz. \n Will YOU go to Ankara tomorrow?

or, ~~~~ sdparse Siz mi yarın Ankara’ya gideceksiniz. \n Will YOU go to Ankara tomorrow? ~~~~

In these cases, we attach the question particle to the word/phrase it modifies, not to the predicate. Currently, we keep the same POS tag and relation label.

Above covers the use of -mI as question particle. -mI may also be used as a subordinating conjunction as in Okulu bitirdin mi işin hazız ‘when/if you finish the school, your job is waiting for you’. In this case, we treat it like a SCONJ introducing a conditional adverbial clause. It is attached to the head of the clause with mark.

Another use of -mI is in emphatic reduplication, such as güzel mi güzel ‘very beautiful’, hızlı mı hızlı ‘very fast’. In these cases it is attached to the preceding word with relation advmod:emph.

Person names and honorifics

The person names are linked with name relation, and headed by the last part of the name.

Turkish honorifics normally follow the (first) name, Ahmet Bey ‘Mr. Ahmet’, Necla Hanım ‘Ms. Necla’, Ayşe öğretmen ‘Teacher Ayşe’, Mustafa Kemal Paşa ‘general Mustafa Kemal’. A few others introduced later, Sayın/Bay/Bayan, precede the last name or first and last name. We use the relation compound and always mark the proper name as the head of the compound (although it is tempting to mark the title as head in the first case).

Noun-verb compounds / light verbs

A number of verbs in Turkish combine with nouns or adjectives to form predicates. Most common verbs used in constructions are verbs, et-, yap- and ol-, but others such as gel-, dur-, kal-, çık-, düş-, çek- may be used to form compound verbs. The usage of these verbs is not restricted to this ‘light’ usage, they may also be used as normal (heavy) verbs. For example, çek- ‘pull’ is used as a light verb in fotoğraf çek- ‘take photo’, but as a ‘heavy’ verb in saç çek- ‘pull hair’.

The copula/auxiliary ol- should only marked as part of a light-verb compound in clear lexicalized cases, e.g., sebep ol- ‘cause’ but not hasta ol- ‘be/become ill’.

The relation between a light verb and the noun it modifies noun is compound. The noun should be marked as the head of the compound.

Ali Ahmet'e yardım etti. \n Ali helped ahmet
compound(yardım, etti)
nsubj(yardim, Ali)
dobj(yardim, Ahmet)


If numbers are spelled out completely, we use a combination of conj and coord to mark it in a head-final fashion.

iki yüz otuz üç 
comp(yüz, iki)
conj(üç, otuz)
conj(üç, yüz)