New Functions of FrameSQL for Multilingual FrameNets
Contents
1. CaS ee Figure 7 Spanish aproximarse and English cpproach psi jO1 01 lou jO1 02 07 jO1 o JO jou o1 01 01 OL a a FrameSQL uses sense description in a Spanish to English electronic dictionary for making a list of corresponding LUs For example the Spanish Adorning frame which involves a static relationship between a Location and a Theme has the LU adornar V whose sense is described in the dictionary as 3 3 tr to adorn beautify grace decorate embellish deck garnish English LUs in the same frame are as follows 4 adorn V blanket V cloak V coat V cover V deck V decorate V dot V encircle V envelop V festoon V fill V film V line V pave V stud V wreathe V 760 When you extract English words which occur both in 3 and 4 you get the English translation equivalents adorn decorate and deck for the Spanish LU adornar I wrote a Perl script to make English translation equivalents for every Spanish LU in this manner and made such a table as Table 1 which is used for comparing semantic structures of corresponding LUs as in Figure 7 FrameSQL provides similar bilingual tables for the JFN and SALSA data TN o fo OO C a ros fw Table 1 Corresponding Spanish and English LUs oF Table 1 shows correspondences between Spanish LUs and English LUs in the Adorning frame The symbol means th2. Attack headX Attempt headN yemot suas BEN SEN IFN 07 YE 0 1 182 Seit Ende letzten Jahres kommen t glich pheme gt guGReEN OBS eit ee ete ee eines themes Wael Lesource AE Hannover lt Goabiemeeten um die autonome Szene wegen des Verdachts der Gr ndung Mitgliedschaft oder Unterst tzung einer terroristischen Vereinigung nach Paragraph 129 a Strafgesetzbuch auszuforschen Translation Figure 5 SALSA Frame Search Mode A O Instituto Cervantes Servicio de tradomatica Resultado de la traducci n Resultado de la traduccion Texto original Estando cerca de Guacin una moto se aproxim por detr s hasta su veh culo Texto traducido Being close of Guacin one moto came near by behind until its vehicle v EEO eee 1 163013 Estando cerca de Guacin cTheme gt titbetece se aproxim 8 lt source MEE cao EREN Translation 9e Figure 6 Translating Spanish to English 4 Comparing Corresponding LUs and Semantic Structures Another search mode of FrameSQL makes it possible to compare semantic structures of corresponding LUs of two languages In this mode FrameSQL shows a table of FEs assigned to annotated sentences For example Figure 7 shows how FEs of the Arriving frame are assigned to annotated sentences of the Spanish LU aproximarse V and its English translation equivalent approach V The numbers on the right in the table sh
3. FN and JFN use the same XML format as BFN but SALSA uses a different XML format for the German database Burchardt et al 2006 I wrote a Perl script to convert the SALSA XML format into that used in BFN and imported the SALSA data into FrameSQL Figure 5 shows the search menu of Arriving frame of the SALSA data which corresponds to Figures 2 3 and 4 Those pages are mutually linked to each other with the hyperlinks BFN SFN SALSA and JFN FrameSQL adds translation buttons to the end of each annotated sentence for translating the Spanish sentence into English When you click on the translation button 1 in Figure 3 FrameSQL sends the Spanish sentence to the Spanish to English machine translation website Instituoto Cervantes http 62 97 114 150 ats 0001 aspx and shows the translation as in Figure 6 FrameSQL provides similar translation buttons for the JFN and SALSA data e090 frame arr FrameSQL Frame Search Menu for Berkeley FrameNet texunit Frame Global FullText Frame Rel FERel Q amp A Gloss Lexical Units Frame Elements Arranging Arrest POS anfahren V Goal Arriving ar a w ankommen V Mode_of_transportation Arson ortecdisplay 89 eintreffen V Path Artificiality Phrase Type E erreichen V Source Gram Func Ta hereinkommen V Theme Assessing Sram unc _ kommen V Assistance Metaphor E Atonement Display limit 400 T5 Attaching
4. FrameNet and FrameSQL In Proceedings of Building Lexical Resources from Semantically Annotated Corpora pp 13 16 European Language Resources Association
5. He drastically changed the original FN data to make this search tool If you have any double about search results you should always refer to and trust the original FN data Figure 1 FrameSQL LexUnit Search Mode You can select another search mode by clicking a button on the upper pane When you click on the Frame button on the upper pane you can select the Frame search mode where you can search and view annotated sentences of a single frame at a time FrameSQL uses lists of frames LUs and FEs as search parameters Figure 2 shows the Frame search mode where a user selected Arriving in the frame list of the middle left pane approach V in the LU list and Source a general direction from which Theme moves in the FE list of the middle right pane When the user clicks on the Search button FrameSQL translates these search parameters into SQL command 1 and executes the search 1 SELECT bncid word0 sentid sentence list frameID list frame list lemma list pos sensedesc word0 fe pt gf headX headN FROM list word0 sentence WHERE word0 fe Source AND pos V AND lemma approach AND list frame Arriving AND list listid sentence listid AND word0 sentid sentence sentid LIMIT 400 In the lower pane of Figure 2 FrameSQL shows search results which include all the example sentences with the FE Source of the LU approach V in the Arriving frame of the BFN data ANDO FrameSQL Frame Search
6. Menu for Berkeley FrameNet frame arr CGexunit Frame Global Fulltext Frame Rel FERel Q8A Search Reset Gloss Lexical Units Frame Elements Arranging 7 Arrest POS ry approach N Cotheme Arriving ae E E approach V Depictive Arson OTEAISPIGY 5 larrival N Goal Artificiality Phrase Type W arrive V Goal_conditions pees Gram Func WF amma a Assessmg Other Tags crest eans Assistance descend_ on V Mode_of_transportation Atonement H enter V Path Attaching Display limit 400 l entrance N Period_of_iterations jgetV Source Attack r make V 1 Theme Attempt headX reach V gt Time Attempt suas headN Attention v PANS SFN SALSA JEN ReadMe FrameSQL 10 SPF 998345 lt Theme gt MAA Eo E ee fel eB Eara TE Pay Sumo E E E E EnEn above all Otto Bauer and Karl Renner also approached Tgt lt Goal gt i lt source gt eset Metaphor Translation 1 1 Figure 2 BFN Frame Search Mode 3 Mutual Hyperlinks of the Same Frame At the left bottom of the middle right pane in Figure 2 there are three hyperlinks SEN SALSA JEN They are linked to FrameSQL search menus of SFN SALSA and JFN of the Arriving frame respectively Since all the three projects of SFN JFN and SALSA adopt the basic notions of BFN FrameSQL search menus for these projects are basically the same Once you get accustomed to one of those search menus you can search those lexical databases
7. New Functions of FrameSQL for Multilingual FrameNets Hiroaki SATO Senshu University Office 8411 2 1 1 Higashimita Tama ku Kawasaki shi Kanagawa ken 214 8580 JAPAN E mail hiroaki isc senshu u ac up Abstract The Berkeley FrameNet Project BFN is making an English lexical database called FrameNet which describes syntactic and semantic properties of an English lexicon extracted from large electronic text corpora Baker et al 1998 Other projects dealing with Spanish German and Japanese follow a similar approach and annotate large corpora FrameSQL is a web based application developed by the author and it allows the user to search the BFN database in a variety of ways Sato 2003 FrameSQL shows a clear view of the headword s grammar and combinatorial properties offered by the FrameNet database FrameSQL has been developing and new functions were implemented for processing the Spanish FrameNet data Subirats and Sato 2004 FrameSQL is also in the process of incorporating the data of the Japanese FrameNet Project Ohara et al 2003 and that of the Saarbr cken Lexical Semantics Acquisition Project Erk et al 2003 into the database and will offer the same user interface for searching these lexical data This paper describes new functions of FrameSQL showing how FrameSQL deals with the lexical data of English Spanish Japanese and German seamlessly 1 FrameSQL is a web based application which I have developed and been maintain
8. at the Perl script cannot find English LUs corresponding to Spanish ones when it compares sense description in the electronic dictionary and the English LUs of the Adorning frame of the BFN data These gaps are common not only in the Adorning frame but also in other frames Human translators could fill appropriate English LUs in some of the gaps when there are corresponding LUs There are also gaps in frame correspondences BFN makes an inventory of frames and FEs for annotating English sentences Most of them work for annotating sentences of other languages because frames are meant to characterize conceptual structures at a basic level of description Some frames however seem to be language specific For example SFN created new frames which are not defined in BFN such as Become different Becoming wet and Being influenced because these are needed for annotating Spanish sentences FrameSQL does not show hyperlinks in search menus of different languages explained in Section 3 when there are not corresponding search menus of the same frame In other words the presence of the hyperlinks in Figures 2 3 4 and 5 shows that the Arriving frames exist not only in BFN but also in SFN JFN and SALSA Gaps of LUs and frames 6 The process of annotation 1s more complicated in FrameNet than in most of other lexical projects because the number of frames and FEs is large Although BFN Training Annotators 761 compiled a user manual w
9. corpus There are hundreds of FE names used in the BFN corpus and showing all the FEs in the pull down menu would confuse the trainee The quiz pages only show those FEs which are used in annotated sentences of a single frame For example each pull down menu in Figure 8 has 12 FEs Cotheme Depictive Goal Goal conditions Manner Means Mode_of transportation Path Period of iterations Source Theme and Time These FEs are used in annotated sentences of the Arriving frame and their definitions appear in the lower pane when you select them in the pull down menu FrameSQL has similar quiz pages for the SFN JFN and SALSA data 7 Conclusion There are lots of search tools to process XML data but only FrameSQL can investigate BFN SFN JFN and SALSA databases seamlessly Mutual hyperlinks among these databases and the new search mode make it easy to compare the semantic structures of corresponding LUs across languages and it could be useful for building multilingual lexical resources or more generally for multilingual studies Training good annotators for FrameNet is difficult and the quiz pages of FrameSQL will be useful for this purpose In the poster presentation with demo session I will show how the functions of FrameSQL mentioned in this paper really work References Baker C F Fillmore C J and Lowe J B 1998 The Berkeley FrameNet project In Proceedings of the COLING ACL pp 86 90 Montreal Canada Burc
10. eNet is based on the theory of Frame Semantics The central idea of the theory is that word meanings must be described in semantic frames which schematically represent conceptual structures Each semantic frame has a group of lexical units LUs A lexical unit 1s a pairing of a word with a sense Cruse 1986 whose semantic properties are described with semantic roles called frame elements FEs For example the Arriving frame has a Basic Searches 758 list of LUs such as approach N approach V arrive N and arrive V and semantic properties of example sentences in this frame are described with its own set of FEs such as Theme the object that moves and Goal the place where Theme ends up There are several search modes in FrameSQL Figure 1 shows the LexUnit search mode http sato fm senshu u ac jp fn2_13 notes index2 html for the BFN data where you can search and view annotated sentences of a single LU at a time ea00 FrameSQL Lexical Unit Search Menu for Berkeley FrameNet LexUnit Frame Global FullText Frame Rel FE Rel Q amp A lemma iece 1 Use the latest version of Safari Firefox or Netscape Navigator 2 Close the other windows or tabs of your browser abandon 3 Turn on JavaScript and Cookie abash 4 Don t block pop up windows abashed QO 4 gt Oct 28th 2006 Prof SATO Hiroaki at Senshu University compiled the new FrameSQL from FN2 Data Release 1 3
11. hardt A Erk K Frank A Kowalski A Pado S Pinkal M 2006 The SALSA Corpus a German Corpus Resource for Lexical Semantics In Proceedings of the Fifth International Conference on Language Resources and Evaluation pp 969 974 European Language Resources Association Cruse D A 1986 Lexical Semantics Cambridge Cambridge University Press Erk K Kowalski A Pinkal M 2003 A Corpus Resource for Lexical Semantics In Proceedings of the Fifth International Workshop on Computational Semantics IWCS pp 106 121 Tilburg The Netherlands Fillmore C J 1976 Frame Semantics and the Nature of Language In Annals of the New York Academy of Sciences Conference on the Origin and Development of Language and Speech Volume 280 pp 20 32 Fillmore C J Baker C F Sato H 2002 The FrameNet Database and Software Tools In Proceedings of the Third International Conference on Language Resources and Evaluation pp 1157 1160 European Language Resources Association Ohara K H Fujii S Saito H Ishizaki S Ohori T Suzuki R 2003 The Japanese FrameNet Project A Preliminary Report In Proceedings of Pacific Association for Computational Linguistics PACLING 03 pp 249 254 Halifax Canada Sato H 2003 FrameSQL A Software Tool for FrameNet In ASIALEX 03 Tokyo Proceedings pp 762 251 258 Asian Association of Lexicography Tokyo Japan Subirats C Sato H 2004 Spanish
12. hich explains every concept of BFN in detail it is still difficult to annotate sentences with rich semantic information It is useful for annotators to exercise annotation on training materials before they start real annotation FrameSQL has quiz pages in which you can try quasi annotation You are asked to select the appropriate FE in the pull down menu for each of phrases Figure 8 shows the quiz page for the LU approach V of the Arriving frame of BFN in which a user is about to select Path in the pull down menu as the FE assigned to the word it in the sentence After the selection FrameSQL automatically marks the user s choice as in Figure 9 where the striked through Path signifies that the choice is the wrong answer The quiz pages also show the definition of the FE which the user selects in the lower pane as in Figure 9 when the user makes a choice in the pull down menu FrameSQL Quiz Quiz Quiz2 0 0O00 LexUnit POS Frame appro Home other Links A n E 0 arriving approach V COD come near or nearer to in distance time approach N epee wer 1 Gietheme ig The few Marxists who devoted serious ee attention to the problem of nationality and pee above all arriving Otto Bauer and Karl Renner also approached e v2 it G H from a class 2a Cotheme me approbation N 56 Debictive judgment 2c Goal L Tet i 2d Goal_conditions Daniel appr
13. in the same manner and compare LUs of the same frame among those languages For example when you click on SFN of Figure 2 the corresponding SFN search menu of the Arriving frame appears in the middle right pane as in Figure 3 The user interface in Figure 3 is basically the same as that of Figure 2 You can execute a search command similar to that in Figure 2 by selecting the Spanish LU aproximarse V to approach and the FE Source of this search menu FrameSQL shows search results in the lower pane where all the example sentences for the LU aproximarse V with the FE Source in the Arriving frame of SFN are displayed FrameSQL is Unicode compliant and it can process 759 Japanese Unicode data of JFN as well as English and Spanish data Figure 4 shows the search menu of Arriving frame for the JFN data A user executed a search after selecting the Japanese LU 7 l V to arrive of the LU list and the FE Source in the FE list in the middle right pane Search results in the lower pane display a Japanese annotated sentence with the FE Source of the Arriving frame 00o frame arr FrameSQL Frame Search Menu for Berkeley FrameNet texunit Frame Global FullText Frame Rel FE Rel Q amp A Arraignment Arranging Search Reset Gloss Lexical Units Frame Elements Manner_of_motion Arrest POS ry acercarse V Means Arriving ee ae a acudir V Medium Arson ortkdisplay ___ 1 adentrar
14. ing for BFN since 1999 The application is designed to search the FrameNet database a lexicon of contemporary written English that the project is producing for both computational and lexicographic interests Other projects dealing with Spanish German and Japanese follow a similar approach and annotate large corpora The Spanish FrameNet Project SFN and the Japanese FrameNet Project JFN have been working within the framework of Frame Semantics Fillmore 1976 which BFN is based on The Saarbr cken Lexical Semantics Acquisition Project SALSA also adopts the paradigm of Frame Semantics and created a large German annotated corpus FrameSQL can search and view the BFN data release 1 3 on a standard web browser You do not need to install any additional software tools to use FrameSQL nor do you even need to download the BFN data to your local computer because FrameSQL accesses the database of the server computer in Japan and executes searches The server computer handles the BFN data with MySQL a popular relational database application that can execute complex searches Directly searching a MySQL database requires learning the database query language SQL but with FrameSQL you do not have to learn SQL at all You only select several search parameters and a form of the output search results on the web browser and FrameSQL translates these parameters into their corresponding SQL command and executes the search Introduction Zs Fram
15. oached 2e Manner WY approving A 2f Means ery good a ee 2g Mode_of_transportation 100 Right 1 Wrong 0 e 2i Period_of_iterations oe 2j Source k rheme Semantic Type wise ir but it need not be 21 Time Physical_object APPROA I ducked as M NZS oN my head 1318 the house Figure 8 Quiz for training annotators 000 LexUnit POS Frame appro FrameSQL Quiz Quiz Quiz2 Home other Links B approach N arriving e approach V arriving approbation N judgment a approving A judgment gt Path arriving approach V COD come near or nearer to in distance time 0 or standard 1 Gictheme jg The few Marxists who devoted serious attention to the problem of nationality and nationalism above all Otto Bauer and Karl Renner also approached it G ig from a class 2h Path 0 standpoint Answers 2 G 00 02 19 Wrong Your average score 50 Right 1 Wrong 1 Your choices 1 1 Theme 1 2 Path ee is any description of a trajectory of motion which is neither a nor a Goal In this frame Path expressions almost always have a Via sense The officer Agate tesi the house Anr tanu CSF z Daniel approached Figure 9 Quiz for training annotators I wrote a Perl script to make the quiz pages from the BFN XML data In other words sentences in the quiz pages are actually used in the BFN
16. ow occurrences of example sentences annotated with those FEs on the left For example the third row Goal Manner Theme 01 01 shows that the LU aproximarse V has one annotated Spanish sentence with the FEs Goal Manner Theme and approach V has one English sentence with the same FEs The numbers on the right are linked with the corresponding annotated sentences and you can access English and Spanish annotated sentences by clicking on the hyperlinked numbers When you click on the hyperlinked number 0 of approach beside Goal Manner Theme FrameSQL will display on your web browser sentence 2 which contains this FE set 2 Guided by the agent tpemesthe pair approached eManner gt With extreme caution using the trees for cover until they emerged from the tree line above the valley LeGoa PNI The symbol lt Goar DNI in 2 corresponds to the parenthesized Goal of Figure 7 The parentheses of Goal signify that the FE Goal is conceptually salient but does not appear in the sentence BFN calls this type of FE null instantiation There are three types of null instantiation and one of them is DNI Definite Null Instantiation in which the missing element is already understood in the linguistic or discourse context FrameSQL parenthesizes all null instantiated FEs aproximarse perce Depictive RRM VEEE Mae Circumstances JeReN VESE GRE SE Themegy eres Sees ee Ge
17. se V Mode_of_transportation Artificiality Phrase Type T aproximarse V Path Assessing Gram Func we arrimarse V Place i A avanzar V Purpose Assistance Metaphor p encaminarse V Re_encoding Atonement Display limit 400 entrar V Reason Attaching inmigrar V Result Attack headX irrumpir V Simultaneous_event cy llegada N 1 Source Attempt headN llegar V Speed Attempt suasy O lt i gt BFN SALSA JEN 01 See 1 163013 Estando cerca de Guacin Themes se aproxim Ppor detr sl lt Goa b ESAR AOne Translation 1 2 3 Figure 3 SFN Frame Search Mode e090 frame arr FrameSQL Frame Search Menu for Berkeley FrameNet texunit Frame Global FullText Frame Rel FERel Q amp A Arraignment Gloss Lexical Units Frame Elements Search Reset Arranging Arrest POS cs t amp LYOND lt N Cotheme Arriving a Ad V Depictive Arson Sort amp display __ RSV Goal Artificiality Phrase Type B ZZV Manner ing Gram Func Means Assessin M h gt Mode_of_transportation Assistance etaphor Path Atonement Display limit 400 5 Source Attaching Theme Attack headX Time Attempt headN Attempt suasic 4 BFN SFN SALSA 01 OSE 5 REE C1 AS EEE Ng AEN Goa laa a Tet Mic EN lOD DTI Do Lethemes DME Translation 1 am S Figure 4 JFN Frame Search Mode S
Download Pdf Manuals
Related Search