Evaluating Complement-Modifier Distinctions in a Semantically - - PowerPoint PPT Presentation

evaluating complement modifier distinctions in a
SMART_READER_LITE
LIVE PREVIEW

Evaluating Complement-Modifier Distinctions in a Semantically - - PowerPoint PPT Presentation

Evaluating Complement-Modifier Distinctions in a Semantically Annotated Corpus Mark McConville and Myroslava O. Dzikovska The Sixth Language Resources and Evaluation Conference (LREC08) Marrakech, 28 May 2008 McConville/Dzikovska Evaluating


slide-1
SLIDE 1

Evaluating Complement-Modifier Distinctions in a Semantically Annotated Corpus

Mark McConville and Myroslava O. Dzikovska

The Sixth Language Resources and Evaluation Conference (LREC’08) Marrakech, 28 May 2008

McConville/Dzikovska Evaluating Complement-Modifier Distinctions LREC’08

slide-2
SLIDE 2

1

The FrameNet corpus

Overshadowed by Grigorovich, Kokonin nonetheless apparently eclipsed him in power in recent months. Kokonin eclipsed him in power in recent months Surpassing Item Standard Attribute Time NP V NP PP PP Ext Obj Dep Dep

McConville/Dzikovska Evaluating Complement-Modifier Distinctions LREC’08

slide-3
SLIDE 3

2

Harvesting a verb lexicon

Kokonin eclipsed him in power in recent months Surpassing Item Standard Attribute Time NP V NP PP PP Ext Obj Dep Dep

❙ ❙ ❙ ❙ ❙ ❙ ❙ ❙ ❙ ❙ ✇ ❇ ❇ ❇ ❇ ❇ ❇ ❇ ❇ ❇ ❇ ❇ ◆ ✄ ✄ ✄ ✄ ✄ ✄ ✄ ✄ ✄ ✄ ✄ ✎ ✚ ✚ ✚ ✚ ✚ ✚ ✚ ✚ ✚ ✚ ❂ ❄

          

  • rth

eclipse syncat V semtype Surpassing args * 2 6 6 6 4 gr Ext cat NP role Item 3 7 7 7 5, 2 6 6 6 4 gr Obj cat NP role Standard 3 7 7 7 5, 2 6 6 6 4 gr Dep cat PP role Attribute 3 7 7 7 5, 2 6 6 6 4 gr Dep cat PP role Time 3 7 7 7 5 +

          

McConville/Dzikovska Evaluating Complement-Modifier Distinctions LREC’08

slide-4
SLIDE 4

3

Removing non-Core arguments

      

  • rth

eclipse semtype Surpassing args * 2 6 6 4 gr Ext cat NP role Item 3 7 7 5, 2 6 6 4 gr Obj cat NP role Standard 3 7 7 5, 2 6 6 4 gr Dep cat PP role Attribute 3 7 7 5, 2 6 6 4 gr Dep cat PP role Time 3 7 7 5 +

             

  • rth

eclipse semtype Surpassing args * 2 6 6 4 gr Ext cat NP role Item 3 7 7 5, 2 6 6 4 gr Obj cat NP role Standard 3 7 7 5, 2 6 6 4 gr Dep cat PP role Attribute 3 7 7 5 +

      

McConville/Dzikovska Evaluating Complement-Modifier Distinctions LREC’08

slide-5
SLIDE 5

4

Question

Does FrameNet’s notion of semantic ‘coreness’ correlate with syntactic complementhood?

McConville/Dzikovska Evaluating Complement-Modifier Distinctions LREC’08

slide-6
SLIDE 6

5

Method 1

verb in VerbNet? 0.95

✑ ✑ ✑ ✑ ✑ ✑ ✑ ✑ ✑ ✰

Y

❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❘

N PP in VerbNet? 0.97

✑ ✑ ✑ ✑ ✑ ✑ ✑ ✑ ✑ ✰

Y

❅ ❅ ❅ ❅ ❅ ❅ ❘

N ignore complement modifier

McConville/Dzikovska Evaluating Complement-Modifier Distinctions LREC’08

slide-7
SLIDE 7

6

Results 1

Core non-Core complements 199 37 non-complements 82 115

Agreement: 0.73 Kappa: 0.65

McConville/Dzikovska Evaluating Complement-Modifier Distinctions LREC’08

slide-8
SLIDE 8

7

Method 2

verb in VerbNet?

✑ ✑ ✑ ✑ ✑ ✑ ✑ ✑ ✑ ✰

Y

❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❘

N PP in VerbNet?

✑ ✑ ✑ ✑ ✑ ✑ ✑ ✑ ✑ ✰

Y

❅ ❅ ❅ ❅ ❅ ❅ ❘

N ignore complement PP in VerbNet+? 0.80

❈ ❈ ❈ ❈ ❈ ❈ ❈ ❈ ❲

Y

❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍ ❍ ❥

N complement modifier

McConville/Dzikovska Evaluating Complement-Modifier Distinctions LREC’08

slide-9
SLIDE 9

8

Results 2

Core non-Core complements 258 49 non-complements 23 103

Agreement: 0.83 Kappa: 0.75

McConville/Dzikovska Evaluating Complement-Modifier Distinctions LREC’08

slide-10
SLIDE 10

9

Method 3

verb in VerbNet?

✑ ✑ ✑ ✑ ✑ ✑ ✑ ✑ ✑ ✰

Y

❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❅ ❘

N PP in VerbNet?

✑ ✑ ✑ ✑ ✑ ✑ ✑ ✑ ✑ ✰

Y

❅ ❅ ❅ ❅ ❅ ❅ ❘

N PP in VerbNet+? 0.94

✑ ✑ ✑ ✑ ✑ ✑ ✑ ✑ ✑ ✑ ✑ ✑ ✑ ✑ ✑ ✑ ✰

Y

✄ ✄ ✄ ✄ ✄ ✄ ✄ ✄ ✄ ✄ ✄ ✎

N complement PP in VerbNet+?

❈ ❈ ❈ ❈ ❈ ❈ ❈ ❈ ❲

Y

❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍ ❍ ❥

N complement modifier

McConville/Dzikovska Evaluating Complement-Modifier Distinctions LREC’08

slide-11
SLIDE 11

10

Results 3

Core non-Core complements 395 59 non-complements 37 145

Agreement: 0.85 Kappa: 0.65

McConville/Dzikovska Evaluating Complement-Modifier Distinctions LREC’08

slide-12
SLIDE 12

11

Core dependents which are not complements

She unfastened [the waistband]Fastener [of her skirt]Containing object

McConville/Dzikovska Evaluating Complement-Modifier Distinctions LREC’08

slide-13
SLIDE 13

12

Conclusions

If we assume that Core = complement:

  • 13% of PP complements will be lost
  • 9% of PPs left will be non-complements

McConville/Dzikovska Evaluating Complement-Modifier Distinctions LREC’08