Kwizifundo ze machine learning kukho iindlela ezimbini zokufunda okanye ukwenza umsebenzi ngee-agents. Eyokuqala yi-competition, enye yi-cooperation. Kwi-science sifunda ngokusebenzisana - umsebenzi womntu uxhomekeke kumsebenzi wonye umntu kodwa abantu bayakhuphisana kancinci. Ngamanye amazwi, idibanisa iindlela ezimbini. Kukho izifundo ezenze imikhuphiswano kuba zifuna indlela yokuphucula ulwazi. Umzekelo, kwi-robotics kukho ukhuphiswano olaziwa ngokuba yi-RoboCup. Olukhuphiswano laqalwa ngonyaka ka 1997, umsebenzi walo aluphelelanga ekuthelekiseni iincutshe, enye into kukutsala abantu abangazi nto ngezizifundo ukuba bazi banzi okanye bafuna ukufunda.

Kwizifundo ze-natural language generation kukho abantu abafuna ukhuphiswano le systems. I-natural language generation kulapho sifunda iindlela zokuguqulela iinkcukacha ezigcinwe ngendlela elungele iikhompyutha, sithatha olu lwazi sikhuphe imibhalo ecacisa olulwazi ngolwimi oluthile. Zininzi iindlela zokwakha ii-systems ezenza le nto, kodwa abantu abaninzi basebenzisa i-architecture eyaziwa ngokuba yi-pipeline. Kubalulekile uyazi ukuba zininzi izinto abantu abangagqinelaniyo ngazo. Yiyo lo nto kubanzima ukwenza ukhuphiswano lwe NLG systems.

Into enomtsalane ngoku lukhuphiswano lwe end-to-end kwi-NLG systems. Le nto ithetha ukuba wonke umntu unikwa i-input enye, i-system egcono ibonwa ngokuthelekiswa kwe-outputs. Ukhuphiswano olunje lungabona kujongwa ii-ouputs kuqwalaselwa izinto ezifana ne:

	accuracy, fluency, etc.

Into entle ngokuphiswano olunje yinto yokuba aluxeli indlela yokukha i-system, le nto ibangela abantu abangenelelayo babe baninzi. Lunazo kodwa iingxaki ukhuphiswano olunje. Singabona ii-systems ezingaqwalaseli ulwimi lonke, ezisombulula iingxaki ezibekwa kukhuphiswano qha. Le nto ingabangela ukuba iincutshe kwezinye izifundo zingayazi okanye zingakuboni ukubaluleka kwezi-systems kunye nomsebenzi wokhuphiswano olu.

Enye yeendlela esingalenza ngalo olukhuphiswano kukuvula ii-systems, sithelekise imiphakathi. Ukwenziwa kwalento kungafana nendlela i-message understanding (MU) yasuka ekuvavanyeni i-task enye ukwenzela ikwazi ukuvavanya ii-tasks ezininzi. Le nto yanyusa inani labantu abakuphisanayo kwi-MU competitions. Ngaphezu koku, yavala isikhewu phakathi kokwenza kakhuhle kukhuphiswano kunye nokuphucula izifundo ze-natural language processing (NLP). Into edingekayo kukwenza le nto kwicala le-NLG. Umehluko kodwa yinto yokuba akukho mali ininzi kumsebenzi we-NLG, ayifani nemali eyayikhona isuka kurhulumente waseMelika kwi-MU (kunye neenkomfa ze message understanding).

Into yokuqala esifuneka siyijonge okanye siyenze kukuvumelana ngee-subtasks ezibalulekileyo kwi-NLG. Phezu Koku kufuneka sivumelane ngee-inputs kunye nee-ouputs zezi subtasks. I-Rags (Reference Architecture for Generation Systems) iphuma emandleni wabantu ababefuna ukwakha i-reference architecture enganceda iincutshe ezakha ii-NLG systems. Abakhi babona ukuba abantu abasebenza kwicala le-NLG abavumelani nge-modules ezidingekayo okanye ekumele ukuba zibekhona kwi-NLG system. Into entle kodwa yinto yokuba bavumelana ngentlobo zolwazi ekumele zibakhona xa usuka kwi-input usiya kwi-ouput. Kubalulekile uyazi ukuba nangona iRAGS yakha i-technology kunye ne systems, ayiphelelanga. Ingxaki yinto yokuba abantu abaninzi bazibona ii-type descriptions zinzima kwaye umntu unyanzeliswa enze izigqibo ezinzima malunga ne-data, ephendula imibuzo efana nokuba ingaba i-data i-conceptual okanye i-semantic?

Bakhona abantu abacinga ukuba iRAGS yafika phambi kwexesha layo. Yiyo lo nto abantu besithi ukuba xa sibheka phambili yenza ingqondo into yokuyidibanisa ne-technology ye-semantic web efana ne-Web ontology language (OWL). Le nto ayithethi ukuba kuzakuba lula ukuvavanya ii-NLG systems (kunye nee-modules), kodwa siyaqonda ukuba izikunyusa umgangatho wee-NLG systems.