3rd Cameleon Workshop, UFRGS Written corpora for human beings. Achille Falaise LIG-GETALP & LIDILEM 1/20

Size: px

Start display at page:

Download "3rd Cameleon Workshop, UFRGS Written corpora for human beings. Achille Falaise LIG-GETALP & LIDILEM 1/20"

Cody Clarke
7 years ago
Views:

1 3rd Cameleon Workshop, UFRGS Written corpora for human beings Achille Falaise LIG-GETALP & LIDILEM 1/20

2 Motivations For lexical studies, language learning, translation, etc. Dictionaries and grammars are not enough... we need context! 2/20

3 Motivations For lexical studies, language learning, translation, etc. Dictionaries and grammars are not enough... we need context! Textual context In this picture, you can see a lot of large rocks on the banks of the river." 3/20

4 Motivations For lexical studies, language learning, translation, etc. Dictionaries and grammars are not enough... we need context! Textual context In this picture, you can see a lot of large rocks on the banks of the river." 4/20

5 Motivations For lexical studies, language learning, translation, etc. Dictionaries and grammars are not enough... we need context! Textual context Syntactic context 5/20

6 Motivations For lexical studies, language learning, translation, etc. Dictionaries and grammars are not enough... we need context! Textual context Syntactic context Structural context 6/20

7 Motivations For lexical studies, language learning, translation, etc. Dictionaries and grammars are not enough... we need context! Textual context Syntactic context Structural context Thematic context Domain, topic, document type, etc. 7/20

8 Consequence : lots of information to store We keep all these informations Text The role of apoptotic mimicry in host-parasite interplay: is death the only alternative for altruistic behavior? 8/20

9 Consequence : lots of information to store We keep all these informations Text Lemmas and POS <TXT>The role of apoptotic mimicry in host-parasite interplay: is death the only alternative for altruistic behavior? </TXT> <tokens> <t i="1" l="the" f="the" c="det" p="d"/> <t i="2" l="role" f="role" c="nom?s" p="n"/> <t i="3" l="of" f="of" c="prep" p="o"/> <t i="4" l="apoptotic" f="apoptotic" c="adj" p="a"/> <t i="5" l="mimicry" f="mimicry" c="nom?s" p="n"/> [...] </tokens> 9/20

10 Consequence : lots of information to store We keep all these informations Text Lemmas and POS Syntactic trees <TXT>The role of apoptotic mimicry in host-parasite interplay: is death the only alternative for altruistic behavior? </TXT> <tokens> <t i="1" l="the" f="the" c="det" p="d"/> <t i="2" l="role" f="role" c="nom?s" p="n"/> <t i="3" l="of" f="of" c="prep" p="o"/> <t i="4" l="apoptotic" f="apoptotic" c="adj" p="a"/> <t i="5" l="mimicry" f="mimicry" c="nom?s" p="n"/> [...] </tokens> <dependances> <g r="det" s="2" c="1"/> <g r="prep" s="2" c="3"/> <g r="nomprep" s="3" c="5"/> <g r="nn" s="5" c="4"/> [ ] </dependances> 10/20

11 Consequence : lots of information to store We keep all these informations Text Lemmas and POS Syntactic trees Structural information <head> <SEQ id="f eti.xml "> <TXT>The role of apoptotic mimicry in host-parasite interplay: is death the only alternative for altruistic behavior? </TXT> <tokens> <t i="1" l="the" f="the" c="det" p="d"/> <t i="2" l="role" f="role" c="nom?s" p="n"/> <t i="3" l="of" f="of" c="prep" p="o"/> <t i="4" l="apoptotic" f="apoptotic" c="adj" p="a"/> <t i="5" l="mimicry" f="mimicry" c="nom?s" p="n"/> [...] </tokens> <dependances> <g r="det" s="2" c="1"/> <g r="prep" s="2" c="3"/> <g r="nomprep" s="3" c="5"/> <g r="nn" s="5" c="4"/> [ ] </dependances> </SEQ> </head> 11/20

12 Consequence : lots of information to store We keep all these informations Text Lemmas and POS Syntactic trees Structural information Document metadata (document type, topic, author, etc.) 12/20

13 But what about the users? Linguists, language learners, translators... Not computer scientists Easily disturbed by complex interfaces Do not read manuals 13/20

14 ScienQuest : corpora for linguists and language learners Spiral development model Advices from users Development Test with users Multiple corpora English scientific texts (15M words) English learners reports (1M words) French scientific texts (5M words) French press (200M words) German press (200M words) Spanish press (200M words) English press (200M words) Russian litterature (500k words) 14/20

15 ScienQuest : corpora for linguists and language learners Demo 15/20

16 ScienQuest : results Since June 2010, 9,021 requests (1,474 sessions) juillet 2010 août 2010 octobre 2010 décembre 2010 février 2011 septembre 2010 novembre 2010 janvier 2011 mars 2011 avril 2011 Nombre de requêtes Nombre de sessions Assisted search : 75% (of which 47% with syntax) Semantic search : 23% Advanced search : 2% 16/20

17 AXiMAG : corpora by and for translators Interactive Multilingual Access Gateways (website translation) 2 aspects : Collaborative (crowdsourcing) aligned copora building Multilingual access to websites 17/20

18 AxiMAG : corpora by and for translators Demo 18/20

19 AXiMAG : future Real world (& real users) test and evaluation ex. LICIA Website Prototype operationalisation AXiMAG society 19/20

20 Thanks! Questions, comments, corpora? /20

SWISS FOOTBALL LEAGUE (Änderungen vorbehalten / sous réserve de modification) (Version: 19.06.2015)

SWISS FOOTBALL LEAGUE (Änderungen vorbehalten / sous réserve de modification) (Version: 19.06.2015) (Änderungen vorbehalten / sous réserve de modification) (Version: 19.06.2015) Saison 2015/2016: Juli / juillet 2015 MI/ME 1. CL Q1 DO/JE 2. EL Q1 FR/VE 3. SA/SA 4. SO/DI 5. MO/LU 6. DI/MA 7. CL Q1 MI/ME