SRL
From Knowitall
Revision as of 19:41, 25 February 2013 by SSoderland (talk | contribs)
More Examples (Feb. 25)
- What do we want to extract from "If ..., then ..." constructions.
- The dependency parse in this example has "If Grandma had wheels" as an advcl modifier to the main clause.
- If Grandma had wheels, she would be a tea trolley.
- have.03 A0=Grandma; A1=wheels
- be.01 AM_ADV=had; A1=she; AM_MOD=would; A2=trolley
- (Grandma, had, wheels) mode:hypothetical
- (she, would be, a tea trolley) context: if (Grandma, had, wheels)
- If Grandma had wheels, she would be a tea trolley.
- The dependency parse in this example has "If Grandma had wheels" as an advcl modifier to the main clause.
- How far to go in extending arg2 with post modifying clauses?
- The dependency graph has appositive link from sedative to drug and we have (Alcohol, is, a drug) -- do we infer the last two tuples?
- Alcohol is a drug, a sedative, which depresses the central nervous system
- be.01 A1=Alcohol; A2=drug
- depress.02 A0=sedative; UnknownRole(R-A0)=which; A1=system
- ? (Alcohol, is, a drug, a sedative, which depresses the central nervous system)
- (Alcohol, is, a drug)
- (a sedative, depresses, the central nervous system)
- ? (a drug, depresses, the central nervous system)
- ? (Alcohol, depresses, the central nervous system)
- Alcohol is a drug, a sedative, which depresses the central nervous system
- The dependency graph has appositive link from sedative to drug and we have (Alcohol, is, a drug) -- do we infer the last two tuples?
- Don't create Time: from every AM_TMP
- Clay holds water well , sometimes perhaps too well
- hold.01 A0=Clay; A1=water; AM_MNR=well; AM_TMP=sometimes
- (Clay, holds, water)
- x (Clay, holds, water) Time: sometimes too well
- Clay holds water well , sometimes perhaps too well
- Here is a new arg role AM-PRR. Treat it like any other arg2.
- The verb with ".LV" seems to be a "light verb" construction, with the noun frame mistake.01
- The frame for become.01 has A2 = "immune to various antibiotics" -- shift the adj to the relation.
- Most people make this mistake and over time can become immune to various antibiotics .
- make.LV A0=people; AM-PRR=mistake
- mistake.01 A0=people; C-V=make
- become.01 A1=people; AM-TMP=over; AM-MOD=can; A2=immune
- (Most people, make, this mistake)
- (Most people, can become immune, to various antibiotics) Time: over time
- Most people make this mistake and over time can become immune to various antibiotics .
Examples of SRL to Extraction Rules
- Create a tuple for every A1 Verb A2 where A1 and A2 are any of {A0, A1, A2, …, A5}
- John was reading a book.
- A0: John read.01 A1: a book
- (John, was reading, a book)
- John sat in the library.
- A1: John sit.01 A2: in the library
- (John, sat, in the library)
- Ignore AM_MNR but use dependency graph to include adverbials in relation
- John was reading quietly.
- A0: John read.01 AM_MNR: quietly
- (John, was reading quietly, )
- John sat quietly in the library.
- A1: John sit.01 AM_MNR: quietly A2: in the library
- (John, sat quietly, in the library)
- Multiple arg2 with same arg1, second arg2 starts with a verb.
- Create two tuples, use both verbs in relation for arg2 that starts with verb.
- Don’t create the tuple with embedded verb.
- John hopes to read the book.
- A0: John hope.01 A1:to read the book
- A0: John read.01 A1: the book
- ? (John, hopes, to read the book)
- (John, hopes to read, the book)
- x (John, to read, the book)
- John hopes to read the book.
- John sat in the library, reading a book.
- A1: John sit.01 A2: in the library AM_PRD: reading a book
- A0: John read.01 A1: a book
- (John, sat, in the library)
- (John, sat reading, a book)
- x(John, reading, a book)
- John sat in the library, reading a book.
- John reads books to stimulate his mind.
- A0: John read.01 A1: books AM_PRP: to stimulate his mind
- A0: John stimulate.01 A1: his mind
- (John, reads, books)
- ? (John, reads books, to stimulate his mind)
- (John, reads to stimulate, his mind)
- x (John, to stimulate, his mind)
- John reads books to stimulate his mind.
- Multiple arg2 with same arg1, second arg2 starts with preposition.
- Append the arg that starts with a preposition to the previous arg2.
- John reads books for stimulating his mind.
- A0: John read.01 A1: books AM_PRP: for stimulating his mind
- A0: John stimulate.01 A1: mind
- (John, reads, books, for stimulating his mind)
- ? (John, reads books, for stimulating his mind)
- Tr (John, reads, books for stimulating his mind)
- Tr (John, reads, books)
- x (John, stimulating, his mind)
- John reads books for stimulating his mind.
- Append the arg that starts with a preposition to the previous arg2.
- John was reading quietly.
- Ignore args that start R-*
- John read a book that discussed philosophy.
- A0: John read.01 A1: a book that discussed philosophy
- A0: a book discuss.01 R-A0: that A1: philosophy
- (John, read, a book that discussed philosophy)
- (a book, discussed, philosophy)
- John read a book that discussed philosophy.
-
- John read a book in which philosophy was discussed.
- A0: John read.01 A1: a book in which philosophy was discussed
- A0: a book discuss.01 R-AM-LOC: in which A1: philosophy
- (John, read, a book in which philosophy was discussed)
- (a book, discussed, philosophy)
- John read a book in which philosophy was discussed.
- Designate the AM_TMP in tuples as Time:
- John read the book last Thursday.
- A0: John read.01 A1: the book AM_TMP: last Thursday
- (John, read, the book) Time: last Thursday
-
- Obama was elected in 2008.
- A0: Obama elect.01 A1: in 2008 AM_TMP: in 2008
- (Obama, was elected, in 2008) Time: in 2008
- Obama was elected in 2008.
- Designate the AM_LOC in tuples as Location:
- John read the book in Paris.
- A0: John read.01 A1: the book AM_LOC: in Paris
- (John, read, the book) Location: in Paris
- John read the book in Paris.
-
- Inslee was elected in Washington State.
- A0: Inslee elect.01 A1: in Washington State AM_LOC: in Washington State
- (Inslee, was elected, in 2008) Location: in Washington State
- Inslee was elected in Washington State.