!!!Notes from meeting 070612 * 1.2 M asks * 1.2. U answers, with * 1.2.1 orthographical error * 1.3.1 correct the orthographical error <==== (210) * run the standard spellchecking test, with a large editing distance (3!), * . delete, add, exchange 1,2,3... times * but on a limited set of words: * a. common answering words for all dialogues * b. dialogue-specific words for each dialogue ** mus lea brlkdajsf ** leago dus bruvsa? ** minulla on konjakkia ** tarkotatko että sinulla on kotikaljaa? Topics * shop * relatives * school * food * drink eventually likely spelling correction if not in set * a á * z ž * etc !!Setting: The machine asks: ''Mitä haluat juoda?'' Then the machine knows that the answer should be with a verb in sg1 and a noun in partitive Any verb spotted should then be analysed. If not in sg1 it should be suggested to be in sg1: Possible answers + corrections: * juoda kahvi juon kahvia * haluta juoda kahvi haluan juoda kahvia * minä haluat juoda kahvin minä haluan juoda kahvia * en otta kahvi en ota kahvia * etc. What we sketch here is actually a restricted grammar checker. 1.2.2 case error ... or any grammatical error in the sentence syntactic templates for the answers * Mitä haluat juoda? * Teetä * Hienoa. (Voitko käyttää kokonaista lausetta?) * Haluan juoda kahvia, mutta ottaisin kuitenkin teetä. * 1.3.2.1 correct case * 1.3.2.2 then more questions on the same case * 1.2.3 correct and forward !theme {{{ iq car a c ra dq > dq > car .. colour .. year model block iq cat question logic from car to cat? }}} !a. global logic = conversation * personal life ** car ** cat ** hobbies * professional life ** work ** studies * non-personal * politics ** ... * shopping ** prices ** colours ** paying * traveling ** where have you been (all place names!!!) ** when travel ** how travel !b. local logic within each subdomain (car, cat, ...) dependent questions !!New setting: Student asks. Give a picture. Instruct the student to ask. Picture of a shop, with objects "Now you are in the shop, find out what these things cost." * Goal. Train a special way of making questions, for a special purpose. * Challenge: Write the grammar of all possible questions * paljonko N-Nom (maksaa/olla+V+Ind+Sg3) * ... * FSIG * S -> NP VP !Analysing * a. check the questions with a fsa (intelligent list) * b. check the questions with CG Generating fsa for generating M instructs S asks M analyses -- is it c !!Tasks * Flow chart of the dialog processing ** different modules and their interaction (Saara) * Flow chart of the dialog process ** different modules and their interaction (Lene) * user: ... maitten ... maiden ... * comp: ... maiden ... (?) * User may use any form from within a relative wide form * Comp should use one form only (Guovdageaidnu?) * User could pick what dialect the comp should use {{{ boy girl adult EAST Ánde Biret .. WEST }}} Gives six parameters, boy-topic, girl-topic, adult-topic for east and west dialect Drawing persons Alpha version in one column, eventually with colour * Normal Black Teacher to the right * C where do you live Black * P I live from Tromsø Grey * T do you mean "in"? * P I live in Tromsø Grey.. * P I live in Tromsø Blach For beta version considering multiple coloums {{{ Normal Black Normal? Grey? Teacher to the right C where do you live P I live from Tromsø T do you mean "in"? T Do you want to practice ? P I live in Tromsø P I live in Tromsø }}} * technical * mastering generation * alpha generator * mastering turn-taking * alpha turntaking machine * mastering analysis * error correction * planning topics * collecting word sets * modeling questions and answers * feedback * what kind of feedback * integrated feedback, ** comment typos, ask user to pick what she intended * independent feedback ** don't remember words? try THESE ones ** don't remember locative? read HERE * other part of the project: visl sentences ** write the sentences ** revise the cg2visl.pl script (has some ps in it) ** write a visl2ps component (treebank projects, conversion visl2tigerxml) * linguistic ** what grammatical categories should be generated ** (go thorough the isme.fst to make it generate unique forms only) ** working on sets !!Time plan * Saara: ** Nothing before holiday ** August: Planning, including correction mechanism ** Autumn: Regular meeting, monthly or so * Lene, Trond ** Autumn: ** Visl: Work ** Look at lærebøker Davvin/Univ.kompendium/Language center´s komp./Johnskareng´s books for primary/sec.school: Kind of questions, word-lists... * Interactive: * Make sets * Make dialogue frames * Make two unique generators (east, west) * Investigate error recognition * Collect corpus of errors??? ** Collect student text (Laila, other sources) ** Make a grammar checker, actually (be able to correct errors) ** Make a template for the answer, and be able to check it. ** "Minä ostan leipä" ** IF Nom-Sg1 TV-Sg1 Nom ** THEN suggest Nom-Sg1 TV-Sg1 Acc ** Flexible disambiguator {{{ Minæ Pron Sg1 näen TV Sg1 tuli N Nom V Sg3 Prt SE Obj IF (0 Acc)(-1 TV); SE Obj IF (0 N)(-1 TV LINK -1 SUBJ)(nowhere else Acc); RE VFIN IF (V)(-1 TV LINK -1 SUBJ); }}}