1. Semantic Enrichment of Text with Background Knowledge Anselmo Peñas NLP & IR Group UNED nlp.uned.es Eduard Hovy USC / ISI isi.edu

2. UNED nlp.uned.es Text omits information San Francisco's Eric Davis intercepted a Steve Walsh pass on the next series to set up a seven-yard Young touchdown pass to Brent Jones.

3. UNED nlp.uned.es Make explicit implicit information Implicit(More) explicit San Francisco’s Eric DavisEric Davis plays for San Francisco E.D. is a player, S.F. is a team Eric Davis intercepted pass 1 - Steve Walsh pass 1 Steve Walsh threw pass 1 Steve Walsh threw interception 1 … Young touchdown pass 2 Young completed pass 2 for touchdown… touchdown pass 2 to Brent JonesBrent Jones caught pass 2 for touchdown San Francisco's Eric Davis intercepted a Steve Walsh pass on the next series to set up a seven-yard Young touchdown pass to Brent Jones.

4. UNED nlp.uned.es Goals General Goal Automatic recovering of such omitted information Enrichment is the process of adding explicitly to a text’s representation the information that is either implicit or missing in the text

5. UNED nlp.uned.es The enrichment cycle Cycle: 1. Read text from collection 2. Ruminate in BKB 3. Enrich text representation 4. Repeat Domain Docs. Reading Background Knowledge Base Rumination Enrichment

6. UNED nlp.uned.es Goals Specific goals of this work Explore the idea of using “Proposition Stores” as Background Knowledge for enrichment Explore procedures for enrichment Determine the kinds of knowledge that Proposition Stores must include to enable enrichment

7. UNED nlp.uned.es Outline 1. Intro 2. BKB 3. Enrichment 4. Features of BKBs for Enrichment 5. Conclusion

8. UNED nlp.uned.es Elements in our BKB Entities Classes: not limited to a predefined set Instances: proper nouns (in this first approach) Class:has-instance:Instance relations Propositions: Predefined syntactic structures NV, NVPN NVN, NVNPN NPN, AN …

9. UNED nlp.uned.es Extraction of propositions Patterns over dependency trees prop( Type, Form : DependencyConstrains : NodeConstrains ). Examples: prop(nv, [N,V] : [V:N:nsubj, not(V:_:'dobj')] : [verb(V)]). prop(nvnpn, [N1,V,N2,P,N3]:[V:N2:'dobj', V:N3:Prep, subj(V,N1)]:[prep(Prep,P)]). prop(has_value, [N,Val]:[N:Val:_]:[nn(N), cd(Val), not(lemma(Val,'one'))]).

10. UNED nlp.uned.es Background Knowledge Base (NFL, US football) ?> NN NNP:’pass’ NN 24 'Marino’:'pass‘ NN 17 'Kelly':'pass' NN 15 'Elway’:'pass’ … ?>X:has-instance:’Marino’ 20 'quarterback':has-instance:'Marino' 6 'passer':has-instance:'Marino' 4 'leader':has-instance:'Marino' 3 'veteran':has-instance:'Marino' 2 'player':has-instance:'Marino' ?> NPN 'pass':X:'touchdown‘ NPN 712 'pass':'for':'touchdown' NPN 24 'pass':'include':'touchdown’ … ?> NVN 'quarterback':X:'pass' NVN 98 'quarterback':'throw':'pass' NVN 27 'quarterback':'complete':'pass‘ … ?> NVNPN 'NNP':X:'pass':Y:'touchdown' NVNPN 189 'NNP':'catch':'pass':'for':'touchdown' NVNPN 26 'NNP':'complete':'pass':'for':'touchdown‘ … ?> NVN 'end':X:'pass‘ NVN 28 'end':'catch':'pass' NVN 6 'end':'drop':'pass‘ …

11. UNED nlp.uned.es Outline 1. Intro 2. BKB 3. Enrichment 4. Features of BKBs for Enrichment 5. Conclusion

12. UNED nlp.uned.es Enrichment example (1) …to set up a 7-yard Young touchdown pass to Brent Jones pass Young touchdown Jones nn to Young pass ?> X:has-instance:Young X=quarterback ?> NVN:quarterback:X:pass X=throw X=complete pass to Jones ?> X:has-instance:Jones X=end ?> NVN:end:X:pass X=catch X=drop

13. UNED nlp.uned.es Enrichment example (2) pass Young touchdown Jones throw complete nn catch drop touchdown pass ?> NVN touchdown:X:pass False ?> NPN pass:X:touchdown X=for …to set up a 7-yard Young touchdown pass to Brent Jones

14. UNED nlp.uned.es Enrichment example (3) pass Young touchdown Jones throw complete for catch drop ?> NVNPN NAME:X:pass:for:touchdown X=complete X=catch …to set up a 7-yard Young touchdown pass to Brent Jones

15. UNED nlp.uned.es Enrichment example (4) pass Young touchdown Jones complete for catch  Young complete pass for touchdown  Jones catch pass for touchdown …to set up a 7-yard Young touchdown pass to Brent Jones

16. UNED nlp.uned.es Enrichment Build context for instances Build context for dependencies Finding prepositions Finding verbs Constrain interpretations

17. UNED nlp.uned.es Enrichment example (5) San Francisco's Eric Davis intercepted a Steve Walsh pass on the next series to set up a seven-yard Young touchdown pass to Brent Jones. Before enrichment for throw catch complete After enrichment

18. UNED nlp.uned.es Outline 1. Intro 2. BKB 3. Enrichment 4. Features of BKBs for Enrichment 5. Conclusion

19. UNED nlp.uned.es What BKBs need for enrichment? (1) Ability to answer about instances Not complete population But allow analogy Ability to constrain interpretations and accumulate evidence Several different queries over the same elements considering different syntactic structures Require normalization (and parsing)

20. UNED nlp.uned.es What BKBs need for enrichment? (1) Ability to discover entity classes with appropriate granularity level Quarterbacks throw passes Ends catch passes Tag an entity as person or even player is not specific enough for enrichment Text frequently introduces the relevant class (appropriate granularity level) for understanding

21. UNED nlp.uned.es What BKBs need for enrichment? (2) Ability to digest enough knowledge adapted to the domain Crucial Approaches Macro-reading (web scale) + domain adaptation Shallow NLP, lack of normalization Reading in context (suggested here) Domain partitioning Deeper NLP, specific domain NLP

22. UNED nlp.uned.es Digest enough knowledge DART: general domain propositions store TextRunner: general domain (web-scale) BKB: specific domain propositions store (only 30,000 docs) ?> quarterback:X:pass DARTTextRunnerBKB (US Football) (no results)(~200) threw (~100) completed (36) to throw (26) has thrown (19) makes (19) has (18) fires (99) throw (25) complete (7) have (5) attempt (5) not-throw (4) toss (3) release

23. UNED nlp.uned.es ?> X:intercept:pass DARTTextRunnerBKB (US Football) (13) person (6) person/place/organization (2) full-back (1) place (30) Early (26) Two plays (24) fumble (20) game (20) ball (17) Defensively (75) person (14) cornerback (11) defense (8) safety (7) group (5) linebacker Digest Knowledge in the domain (entity classes)

24. UNED nlp.uned.es Digest Knowledge in the domain (ambiguity problem) ?> person:X:pass DARTTextRunnerBKB (US Football) (47) make (45) take (36) complete (30) throw (25) let (23) catch (1) make (1) expect (22) gets (17) makes (10) has (10) receives (7) who has (7) must have (6) acting on (6) to catch (6) who buys (5) bought (5) admits (5) gives (824) catch (546) throw (256) complete (136) have (59) intercept (56) drop (39) not-catch (37) not-throw (36) snare (27) toss (23) pick off (20) run

25. UNED nlp.uned.es Domain issue ?> person:X:pass NFL Domain 905:nvn:[person:n, catch:v, pass:n]. 667:nvn:[person:n, throw:v, pass:n]. 286:nvn:[person:n, complete:v, pass:n]. 204:nvnpn:[person:n, catch:v, pass:n, for:in, yard:n]. 85:nvnpn:[person:n, catch:v, pass:n, for:in, touchdown:n]. IC Domain 6:nvn:[person:n, have:v, pass:n] 3:nvn:[person:n, see:v, pass:n] 1:nvnpn:[person:n, wear:v, pass:n, around:in, neck:n] BIO Domain

26. UNED nlp.uned.es Domain issue ?> X:receive:Y NFL Domain 55:nvn:[person:n, receive:v, call:n]. 34:nvn:[person:n, receive:v, offer:n]. 33:nvn:[person:n, receive:v, bonus:n]. 29:nvn:[team:class, receive:v, pick:n]. IC Domain 78 nvn:[person:n, receive:v, call:n] 44 nvn:[person:n, receive:v, letter:n] 35 nvn:[group:n, receive:v, information:n] 31 nvn:[person:n, receive:v, training:n] BIO Domain 24 nvn:[patients:n, receive:v, treatment:n] 14 nvn:[patients:n, receive:v, therapy:n] 13 nvn:[patients:n, receive:v, care:n]

27. UNED nlp.uned.es Outline 1. Intro 2. BKB 3. Enrichment 4. Features of BKBs for Enrichment 5. Conclusion

28. UNED nlp.uned.es Conclusions Limiting to a specific domain provides some powerful benefits Ambiguity is reduced Higher density of relevant propositions Different distribution of propositions across domains Amount of source text is reduced, allowing deeper processing such as parsing Specific tools for specific domains Proposition stores seem to be useful Improve parsing, corref, WSD,… We presented a new application: ENRICHMENT

29. UNED nlp.uned.es Current work Develop automatic procedures for Enrichment Need better Proposition Stores Selectional Preferences Lexical relatedness Structural /frame transformations …

30. UNED nlp.uned.es Future work Develop appropriate methodologies for evaluation Intrinsic? Extrinsic: QA over single documents? Reading comprehension tests?

31. Thanks!

32. UNED nlp.uned.es NVN 3 'quarterback':'find':'receiver‘ NVNPN 3 'quarterback':'throw':'pass':'to':'receiver' NVNPN 2 'quarterback':'complete':'pass':'to':'receiver' NVNPN 1 'receiver':'catch':'pass':'from':'quarterback‘ nvn:('NNP':'quarterback'):'hit':('NNP':'receiver'),177). nvnpn:('NNP':'quarterback'):'throw':'pass':'to':('NNP':'receiver'),143). nvnpn:('NNP':'quarterback'):'complete':'pass':'to':('NNP':'receiver'),79). nvn:('NNP':'quarterback'):'find':('NNP':'receiver'),69). nvnpn:('NNP':'receiver'):'catch':'pass':'from':('NNP':'quarterback'),43).