Towards a German frame based QA System Frame Semantics and QA - - PowerPoint PPT Presentation

Towards a German frame based QA System

Frame Semantics and QA Andrei Beliankou

LDV/CL, Trier

July 28, 2010

Welche Farbe hat eine Maracuja?

What colour is passion fruit?

• A. Beliankou (Trier)

QA July 28, 2010 2 / 26

Outline

1

What is QA? IR and QA: short comparison Definition

2

Approaches to Answering Questions

3

Related work on QA using SR

4

Frame based formalism

5

Tools

• A. Beliankou (Trier)

QA July 28, 2010 3 / 26

What is QA?

Douglas Adams and QA

• A. Beliankou (Trier)

QA July 28, 2010 4 / 26

What is QA?

Douglas Adams and QA

The Ultimate Question of Life, the Universe, and Everything

• A. Beliankou (Trier)

QA July 28, 2010 4 / 26

What is QA?

Douglas Adams and QA

The Ultimate Question of Life, the Universe, and Everything

• A. Beliankou (Trier)

QA July 28, 2010 4 / 26

What is QA? IR and QA: short comparison

A Typical Search Engine

• A. Beliankou (Trier)

QA July 28, 2010 5 / 26

What is QA? IR and QA: short comparison

A Typical Search Engine

• A. Beliankou (Trier)

QA July 28, 2010 6 / 26

What is QA? IR and QA: short comparison

A Typical Search Engine

• A. Beliankou (Trier)

QA July 28, 2010 7 / 26

What is QA? IR and QA: short comparison

Social QA

• A. Beliankou (Trier)

QA July 28, 2010 8 / 26

What is QA? IR and QA: short comparison

Well grounded QA Systems

• A. Beliankou (Trier)

QA July 28, 2010 9 / 26

What is QA? Definition

What is QA?

Question Answering

Question Answering (QA) is an interactive human computer process that encompasses understanding a user information need, typically expressed in a natural language query; retrieving relevant documents, data or knowledge from selected sources; extracting, qualifying and prioritizing available answers from these sources; and presenting and explaining responses in an effective manner. Maybury [2004, p. 3]

• A. Beliankou (Trier)

QA July 28, 2010 10 / 26

What is QA? Definition

What is QA?

Question Answering

Question Answering (QA) is an interactive human computer process that encompasses understanding a user information need, typically expressed in a natural language query; retrieving relevant documents, data or knowledge from selected sources; extracting, qualifying and prioritizing available answers from these sources; and presenting and explaining responses in an effective manner. Maybury [2004, p. 3]

• A. Beliankou (Trier)

QA July 28, 2010 10 / 26

What is QA? Definition

What is QA?

Question Answering

Question Answering (QA) is an interactive human computer process that encompasses understanding a user information need, typically expressed in a natural language query; retrieving relevant documents, data or knowledge from selected sources; extracting, qualifying and prioritizing available answers from these sources; and presenting and explaining responses in an effective manner. Maybury [2004, p. 3] How can we understand questions automatically?

• A. Beliankou (Trier)

QA July 28, 2010 10 / 26

What is QA? Definition

What is QA?

Question Answering

Question Answering (QA) is an interactive human computer process that encompasses understanding a user information need, typically expressed in a natural language query; retrieving relevant documents, data or knowledge from selected sources; extracting, qualifying and prioritizing available answers from these sources; and presenting and explaining responses in an effective manner. Maybury [2004, p. 3] How can we understand questions automatically? What is the technique for extracting relevant information? What should we do if there is no out of the box answer?

• A. Beliankou (Trier)

QA July 28, 2010 10 / 26

What is QA? Definition

What is QA?

Question Answering

Question Answering (QA) is an interactive human computer process that encompasses understanding a user information need, typically expressed in a natural language query; retrieving relevant documents, data or knowledge from selected sources; extracting, qualifying and prioritizing available answers from these sources; and presenting and explaining responses in an effective manner. Maybury [2004, p. 3] How can we understand questions automatically? What is the technique for extracting relevant information? Can we measure the information weight of an answer? What should we do if there is no out of the box answer?

• A. Beliankou (Trier)

QA July 28, 2010 10 / 26

What is QA? Definition

What is QA?

Question Answering

Question Answering (QA) is an interactive human computer process that encompasses understanding a user information need, typically expressed in a natural language query; retrieving relevant documents, data or knowledge from selected sources; extracting, qualifying and prioritizing available answers from these sources; and presenting and explaining responses in an effective manner. Maybury [2004, p. 3] How can we understand questions automatically? What is the technique for extracting relevant information? Can we measure the information weight of an answer? What should we do if there is no out of the box answer? What is the best way to show up the answers?

• A. Beliankou (Trier)

QA July 28, 2010 10 / 26

What is QA? Definition

Prototype of a QA System

• A. Beliankou (Trier)

QA July 28, 2010 11 / 26

Approaches to Answering Questions

Alignment of Question and Answers

Die Shale ist bei den bis hühnereigroßen, kugeligen bis ovalen Früchten der f. edulis braun-violett und wird im Reifezustand runzlig.

• A. Beliankou (Trier)

QA July 28, 2010 12 / 26

Approaches to Answering Questions

Alignment of Question and Answers

Die Shale ist bei den bis hühnereigroßen, kugeligen bis ovalen Früchten der f. edulis braun-violett und wird im Reifezustand runzlig. Welche Farbe hat die Maracuja-Frucht? Wie sieht die Maracuja-Beere aus? Wie ist das Aussehen der Maracuja? Wie ist die Farbe der Maracuja? Was für eine Farbe hat die Maracuja? * Wie erscheint die Maracuja?

• A. Beliankou (Trier)

QA July 28, 2010 12 / 26

Approaches to Answering Questions

Alignment of Question and Answers

Die Shale ist bei den bis hühnereigroßen, kugeligen bis ovalen Früchten der f. edulis braun-violett und wird im Reifezustand runzlig. Welche Farbe hat die Maracuja-Frucht? Wie sieht die Maracuja-Beere aus? Wie ist das Aussehen der Maracuja? Wie ist die Farbe der Maracuja? Was für eine Farbe hat die Maracuja? * Wie erscheint die Maracuja?

Main Question!!!

How do we know, that these questions do correspond to this answer?

• A. Beliankou (Trier)

QA July 28, 2010 12 / 26

Approaches to Answering Questions

Is it an Alignment Problem?

How should we treat "not alignable" tokens? Where can we get large aligned parallel corpora of Questions and Answers?

• A. Beliankou (Trier)

QA July 28, 2010 13 / 26

Approaches to Answering Questions

Is it an Alignment Problem?

How should we treat "not alignable" tokens? Where can we get large aligned parallel corpora of Questions and Answers? Hm, we need some kind of intermediate semantic

• representation. . .
• A. Beliankou (Trier)

QA July 28, 2010 13 / 26

Approaches to Answering Questions

Architecture of a QA System

• A. Beliankou (Trier)

QA July 28, 2010 14 / 26

Related work on QA using SR

Frames

Definition

Frame characterizes a small abstract „scene“ or „situation“, so that to understand the semantic structure of the verb it was necessary to understand the properties of such schematized scenes. [Ruppenhofer et al., 2006] (cf. Tannen [1993] for a good overview) A Theory of the semantic case, developed by Ch. Fillmore [Fillmore, 1985]; Semantic Frames are powerfull enough, but the parser implementation is still relatively easy [Gildea and Jurafsky, 2002, Lluís and Màrquez, 2008]; Nowadays there are many works on SR for QA, but they describe this formalism only for parsing answers and only as an additional technique [Shen and Lapata, 2007, Pizzato and Molla, 2008, Narayanan and Harabagiu, 2004].

• A. Beliankou (Trier)

QA July 28, 2010 15 / 26

Related work on QA using SR

Example of a Frame

Frame Title: Commerce Buy Frame Evoking Element: Kauf Frame Elements (Roles): Buyer, Goods Frame Description: These are words describing a basic commercial transaction involving a buyer and a seller exchanging money and goods, taking the perspecitive of the buyer.

• A. Beliankou (Trier)

QA July 28, 2010 16 / 26

Frame based formalism

Frame based formalism

Figure: from the FEAST Talk by Hagen Fürstenau, Nov 26, 2008

• A. Beliankou (Trier)

QA July 28, 2010 17 / 26

Frame based formalism

Enchanced Architecture

• A. Beliankou (Trier)

QA July 28, 2010 18 / 26

Frame based formalism

Parsing Answers

• A. Beliankou (Trier)

QA July 28, 2010 19 / 26

Frame based formalism

Answering simple Questions

• A. Beliankou (Trier)

QA July 28, 2010 20 / 26

Frame based formalism

Inference based QA

• A. Beliankou (Trier)

QA July 28, 2010 21 / 26

Tools

Tools & Ressources

Software Tools:

YANSER - Yahoo! ANSwers gatherER; SHALMANESER - SHALlow seMANtic parSER [Erk and Padó, 2006]; SALTO - SALsa TOol [Burchardt et al., 2006]; TigerRegistry & TigerSearch [Lezius, 2002]; TreeAligner [Lundborg et al., 2007].

Corpora:

TIGER; FrameNet; SALSA.

• A. Beliankou (Trier)

QA July 28, 2010 22 / 26

Tools

References I

Aljoscha Burchardt, Katrin Erk, Anette Frank, Andrea Kowalski, and Sebastian Padó. SALTO - a versatile multi-level annotation tool. In Proceedings of the 5th International Conference on Language Ressources and Evaluation (LREC 2006), pages 517–520, Genoa, Italy, may 2006. Katrin Erk and Sebastian Padó. Shalmaneser - a toolchain for shallow semantic parsing. In Proceedings of the Fifth International Language Resources and Evaluation (LREC’06), pages 527–532, Italy, Genoa, May

• 2006. European Language Resources Association (ELRA).

Charles J. Fillmore. Frames and the semantics of understanding. Queaderni di Semantica, VI(2):222–254, December 1985. Daniel Gildea and Daniel Jurafsky. Automatic labeling of semantic roles. Computational Linguistics, 28(3):245–288, 2002.

• A. Beliankou (Trier)

QA July 28, 2010 23 / 26

Tools

References II

Wolfgang Lezius. TIGERSearch - ein suchwerkzeug für baumbanken. In Stephan Busemann, editor, Proceedings der 6. Konferenz zur Verarbeitung natürlicher Sprache (KONVENS 2002), pages 107–114, Saarbrücken, 2002. Xavier Lluís and Lluís Màrquez. A joint model for parsing syntactic and semantic dependencies. In CoNLL 2008: Proceedings of the Twelfth Conference on Computational Natural Language Learning, pages 188–192, Manchester, England, August 2008. Coling 2008 Organizing Committee. Joakim Lundborg, Torsten Marek, Mael Mettler, and Martin Volk. Using the stockholm TreeAligner. In Proceedings of The 6th Workshop on Treebanks and Linguistic Theories, Bergen, Norway, December 2007. Mark T. Maybury, editor. New Directions in Question Answering. Mit Press, November 2004.

• A. Beliankou (Trier)

QA July 28, 2010 24 / 26

Tools

References III

Srini Narayanan and Sanda Harabagiu. Question answering based on semantic structures. In Proceedings of the 20th international conference on Computational Linguistics - COLING ’04, pages 693–es, Geneva, Switzerland, 2004. Association for Computational Linguistics. Luiz Augusto Pizzato and Diego Molla. Indexing on semantic roles for question

• answering. In Coling 2008: Proceedings of the 2nd workshop on

Information Retrieval for Question Answering, pages 74–81, Manchester, UK, 2008. Coling 2008 Organizing Committee. Josef Ruppenhofer, Michael Ellsworth, Miriam R. L. Petruck, Christopher R. Johnson, and Jan Scheffczyk. FrameNet II: Extended Theory and Practice. August 2006. URL http://framenet.icsi.berkeley.edu/

index.php?option=com_wrapper&Itemid=126.

• A. Beliankou (Trier)

QA July 28, 2010 25 / 26

Tools

References IV

Dan Shen and Mirella Lapata. Using semantic roles to improve question

• answering. In Proceedings of EMNLP-CoNLL, pages 12–21, Prague, Czech

Republic, 2007. Association for Computational Linguistics. Deborah Tannen. Framing in discourse. Oxford University Press, New York, 1993.

• A. Beliankou (Trier)

QA July 28, 2010 26 / 26