nulls in propbank sep 17 th 2010. what is a null category
TRANSCRIPT
Nulls in PropBank
Sep 17th 2010
What is a null category
PropBank annotation
• Shyam ne khana khaaya• Annotate Arguments of the verb `Khaa’• Shyam and khana
• What if someone asks- kisne khaana khaaya?• Shyam ne khaaya• Shyam ne *khaana* khaaya (Object dropped)
• What about: tumne Ram ko kaunsi kitaab di?• Nili kitaab raam ko di• *maine* nili kitaab raam ko di (Subject dropped)
Missing information!
• Same verb, but varying number of arguments:– Ram ne khaana khaaya– Ram ne *khaana* khaaya
• No consistency in the representation of the verb `khaana’
• Hence, we need to insert missing arguments, or as they are also known ‘Empty Categories (EC)’
Which missing arguments?
• Only the obligatory arguments of the verb• Khaa requires two arguments- eater and thing eaten• When we ask a question: Ram ne kya khaaya?• It becomes possible to drop Ram & respond by saying:
khaana khaaya• Ram is old information and is not repeated
• If obligatory arguments are missing, the correct argument structure of a verb will not be understood by the machine
That’s great!
Four categories of EC
• pro• PRO• GAP-pro• RELPRO
pro
• Missing subject and object e.g responding to a question– *pro* mujhe hari ko denaa hai– Kisne darvaaza kholaa? Mohan ne *pro* kholaa
• Dropped subject in a sentence– Bacchon ke liye karyakram aayojan karna hogaa
pro characteristics
• Fewer surface cues• Requires knowledge of the verb’s obligatory
arguments
pro is the missing VIP!
PRO
• When a sentence has an embedded clause with a non-finite verb AND the same (missing) subject as the main clause, we get *PRO* – Note two important points:– Mohani ne [*PRO*i kitaab paRnii] caahi• [Khaana khaana] has the verb `khaana’ in non-finite
form i.e it’s not showing agreement or tense info• We coindex the PRO with Mohan• There are exceptions!!
For PRO cases, the verb is non-finite, and handicapped!
Exceptions to PRO
• Not present for the modals e.g– Raam ko ghar jaana paDaa / caahiye – Raam ko ghar jaane diyaa
• Verbs with lagnaa (more of a helping verb)– Raam naachne lagaa
• Not in location information type sentences– Ram ghar jaane se pehle vahan rukaa
PRO characteristics
• भा�रत सर्का��र ने [ऐस� र्कारने स] इं र्का�र र्कार दि�या�
The verb ends with ने�, ने�, ने It has a corresponding argument to coindex with! If you replace it with an actual word, it sounds
ungrammatical
GAP PRO
• Only in co-ordinate clauses!!!• Clauses joined by और, या� or a comma
– mohan-nei kitaab paRh-ii aur [*GAP-pro*]i so ga-yaa
– Note that we only look for missing arguments in co-ordination structures
– We also coindex like the *PRO*
GAP-PRO only found with co-ordination!
Context for RELPRO
(1) phool jo daaliyon par khile hue hai, ache hai
• (1) is an example of a relative clause– phool is the noun that’s getting modified– In a relative clause, there’s always an embedded
‘modifier’ clause inside the main clause– phool [jo daaliyon par khile hue hai], ache hai
Upside down clause
• In the case of participial relatives, the order of elements in the relative clause is flipped
• [daaliyon par khile] phool ache hai• The embedded clause is before the element being
modified• Also, there is no relative pronoun ‘jo’ like we can see in (1)
• In such conditions, we will insert a missing subject argument in the upside down clause
• (2) [*RELPRO*daaliyon par khile hue] phool ache hai
RELPRO characteristics
• इंसस [तटों� पर रहने वा�ले] जि�वा � त�ओं र्का� भा� ने�र्कास�ने ह�गा�
Participial verbs can be identified by -त� हुआ , -ने वा�ले e.g र्कारत� हुआ, र्कारने वा�ले
They also have the drel nmod__k1inv or nmod__k2inv
RELPRO: the upside down EC
More practice!