Fahimta da sarrafa maganganun magana babban ƙalubale ne ga jagorancin sarrafa harshe na halitta a cikin mahallin mataimakan AI kamar Amazon Alexa. Wannan matsalar yawanci ta ƙunshi haɗa karin magana daidai a cikin tambayoyin mai amfani tare da ra'ayoyi masu ma'ana, alal misali, kwatanta karin magana "su" a cikin bayanin "kiɗa sabon kundi" tare da wasu masu fasaha na kiɗa. Kwararrun AI a Amazon suna aiki tuƙuru akan fasahar da za su iya taimakawa AI aiwatar da irin waɗannan buƙatun ta hanyar gyare-gyare ta atomatik da sauyawa. Don haka, buƙatar "Kunna sabon kundin su" za a maye gurbinsa ta atomatik tare da "Kunna sabon kundi na Dragons." A wannan yanayin, an zaɓi kalmar da ake buƙata don maye gurbin daidai da tsarin yiwuwar ƙididdigewa ta amfani da koyo na inji.
Masana kimiyya sakamako na farko na aikinsa a cikin tsari mai mahimmanci tare da take mai wuyar gaske - "Scaling state tracking multi- domain dialogue ta amfani da sake fasalin tambaya." Nan gaba kadan, ana shirin gabatar da wannan bincike a reshen Arewacin Amurka na Association for Computational Linguistics.
"Saboda injin ɗinmu na gyaran gyare-gyare yana amfani da ka'idoji na gaba ɗaya don yin amfani da hanyoyin haɗin gwiwar magana, baya dogara da kowane takamaiman bayani game da aikace-aikacen inda za a yi amfani da shi, don haka baya buƙatar sake horarwa lokacin da muke amfani da shi don tsawaita ƙarfin Alexa," in ji shi. Arit Gupta (Arit Gupta), masanin ilimin harshe a Amazon Alexa AI. Ya lura cewa sabuwar fasahar su, da ake kira CQR (sake rubuta tambaya na yanayi), gaba ɗaya yana 'yantar da lambar mataimakan muryar ciki daga duk wata damuwa game da maganganun magana a cikin tambayoyin.
Na farko, AI yana ƙayyade ainihin mahallin buƙatun: menene bayanin mai amfani yake so ya karɓa ko wane mataki zai yi. Yayin tattaunawa tare da mai amfani, AI yana rarraba kalmomi, adana su a cikin masu canji na musamman don ƙarin amfani. Idan buƙatun na gaba ya ƙunshi kowane tunani, AI zai yi ƙoƙarin maye gurbinsa da mafi kusantar kalmomin da aka adana da ma'ana, kuma idan wannan ba a cikin ƙwaƙwalwar ajiya ba, zai juya zuwa ƙamus na ciki na ƙimar da aka fi amfani da su akai-akai. , sa'an nan kuma sake gina buƙatun tare da maye gurbin da aka yi amfani da shi, don mika shi ga mai taimakawa muryar don aiwatarwa.
Kamar yadda Gupta da abokan aiki suka nuna, CQR yana aiki a matsayin riga-kafi don umarnin murya kuma yana mai da hankali kawai akan ma'anar ma'anar kalmomi da ma'ana. A cikin gwaje-gwajen da aka samu horo na musamman, CQR ya inganta daidaiton tambaya da kashi 22% lokacin da mahaɗin da ke cikin tambayar na yanzu yana nufin kalmar da aka yi amfani da ita a cikin amsar da ta gabata, kuma da kashi 25% lokacin da mahaɗin da ke cikin furucin na yanzu yana nufin kalma. daga wata magana da ta gabata.
source: 3dnews.ru