As"deep learning"and"neural network" has become the mainstream technology of natural language processing today, language resources of many small languages in the worldare relatively deficient, and they can not meet the processing pattern based on the "big data".In this case, it is necessary to consider the injection of language knowledgeto expect the realization of the understanding and processing of language at different levels through "deep learning".Therefore, it is decisive to pay attention to the grammatical and semantic characteristics of different languages and sum up the rules for thenatural languageprocessing. Mongolian language is a typical agglutinate language, and its word formation and configuration are all realized by attaching various supplementary components to the stem.The grammatical meaning of a Mongolianword can only be expressed with phrases or sentences in most western languages and oriental languages, and this kind of changes of the words in real Mongolian text constitute about 82% of all words.If these changes are not taken into account and each word is only understood by its stem meaning,it would be impossible to correctly handle the entire text.As a special case of the language knowledge description, we have summarized the grammar rules of a Mongolian wordᠶᠠᠪᠤ .It is worth mentioning that the rulesare only at the lexical change leveland its related partsof one semantic item of the word stem , and the other more semantic items and the relevant rules are not exhaustively described. The Mongolian word "ᠶᠠᠪᠤgo " and its relevantrules ᠶᠠᠪᠤᠭᠤᠯᠵᠠᠭᠠᠴᠢᠬᠠᠵᠠᠢ(They) have let ( all people) go. ---YABV/Ve2+GVL/Fe11+JAGA/Fe5+CIHA/Fi21+JAI/Fs11 ---goYABV/Ve2(stem-imperative form-second person)+GVL/Fe11（causative voice）+JAGA/Fe5（multiple voice）+CIHA/Fi21（perfective aspect）+JAI/Fs11（past tense-statement） ᠶᠠᠪᠤYABV/Ve2 ᠶᠠᠪᠤᠵᠠᠢYABV/Ve2+JAI/Fs11 ᠶᠠᠪᠤᠭᠤᠯ YABV/Ve2+GVL/Fe11 ᠶᠠᠪᠤᠭᠤᠯᠵᠠᠢYABV/Ve2+GVL/Fe11+JAI/Fs11 ᠶᠠᠪᠤᠵᠠᠭᠠ YABV/Ve2+JAG_A/ Fe5 ᠶᠠᠪᠤᠵᠠᠭᠠᠵᠠᠢYABV/Ve2+JAGA/Fe5+JAI/Fs11 ᠶᠠᠪᠤᠴᠢᠬᠠ YABV/Ve2+CIH_A/Fi21 ᠶᠠᠪᠤᠴᠢᠬᠠᠵᠠᠢ?YABV/Ve2+CIHA/Fi21+JAI/Fs11 ᠶᠠᠪᠤᠭᠤᠯᠵᠠᠭᠠ YABV/Ve2+GVL/Fe11+JAG_A/Fe5 ᠶᠠᠪᠤᠭᠤᠯᠵᠠᠭᠠᠵᠠᠢYABV/Ve2+GVL/Fe11+JAGA/Fe5+JAI/Fs11 ᠶᠠᠪᠤᠭᠤᠯᠴᠢᠬᠠYABV/Ve2+GVL/Fe11+CIH_A/Fi21 ᠶᠠᠪᠤᠭᠤᠯᠴᠢᠬᠠᠵᠠᠢYABV/Ve2+GVL/Fe11+CIHA/Fi21+JAI/Fs11 ᠶᠠᠪᠤᠵᠠᠭᠠᠴᠢᠬᠠYABV/Ve2+JAGA/Fe5+CIH_A/Fi21 ᠶᠠᠪᠤᠵᠠᠭᠠᠴᠢᠬᠠᠵᠠᠢYABV/Ve2+JAGA/Fe5+CIHA/Fi21+JAI/Fs11 ᠶᠠᠪᠤᠭᠤᠯᠵᠠᠭᠠᠴᠢᠬᠠYABV/Ve2+GVL/Fe11+JAGA/Fe5+CIH_A/Fi21 ᠶᠠᠪᠤᠭᠤᠯᠵᠠᠭᠠᠴᠢᠬᠠᠵᠠᠢYABV/Ve2+GVL/Fe11+JAGA/Fe5+CIHA/Fi21+JAI/Fs11 The pronouns in co-occurrence with the verbᠶᠠᠪᠤ ᠪᠢ ᠨᠠᠮᠠ ᠪᠢᠳᠡBI/Rb11NAM_A/Rb12 BIDE/Rb13 ᠴᠢ ᠴᠢᠮᠠ ᠲᠠᠨᠠᠷCI/Rb21CIM_A/Rb22 TANAR/Rb23 ᠲᠡᠷᠡ ᠲᠡᠭᠦᠨ ᠲᠡᠳᠡᠨᠡᠷTERE/Rb31TEGUN/Rb32 TEDENER/Rb33 The following is B0 Rule Set (the subject or the agent is the second person singular or plural, or singularplural). ᠴᠢ ᠲᠠᠨᠠᠷ ᠶᠠᠪᠤ CI/Rb21|| TANAR/Rb23→YABV/Ve2 ᠴᠢ ᠲᠠᠨᠠᠷ (ᠨᠠᠮᠠ ᠪᠢᠳᠡ᠂ᠲᠡᠭᠦᠨ ᠲᠡᠳᠡᠨᠡᠷ) ᠶᠠᠪᠤᠭᠤᠯ CI/Rb21 || TANAR/Rb23(NAM_A/Rb12 || BIDE/Rb13 || TEGUN/Rb32 || TEDENER/Rb33) →YABV/Ve2+GVL/Fe11 ᠲᠠᠨᠠᠷ ᠶᠠᠪᠤᠵᠠᠭᠠ TANAR/Rb23→YABV/Ve2+JAG_A/Fi5 ᠴᠢ ᠲᠠᠨᠠᠷ ᠶᠠᠪᠤᠴᠢᠬᠠ CI/Rb21 TANAR/Rb23→YABV/Ve2+CIH_A/Fi21 ᠲᠠᠨᠠᠷ (ᠪᠢᠳᠡ᠂ᠲᠡᠳᠡᠨᠡᠷ) ᠶᠠᠪᠤᠭᠤᠯᠵᠠᠭᠠ TANAR/Rb23(BIDE/Rb13TEDENER/Rb33) →YABV/Ve2+GVL/Fe11+JAG_A/Fe5 ᠴᠢ ᠲᠠᠨᠠᠷ (ᠨᠠᠮᠠᠪᠢᠳᠡ᠂ ᠲᠡᠭᠦᠨᠲᠡᠳᠡᠨᠡᠷ) ᠶᠠᠪᠤᠭᠤᠯᠴᠢᠬᠠ CI/Rb21TANAR/Rb23 (NAM_A/Rb12BIDE/Rb13TEGUN/Rb32TEDENER/Rb33) →YABV/Ve2+GVL/Fe11+CIH_A/Fi21 ᠲᠠᠨᠠᠷ ᠶᠠᠪᠤᠵᠠᠭᠠᠴᠢᠬᠠ TANAR/Rb23 YABV/Ve2+JAGA/Fe5+CIH_A/Fi21 ᠲᠠᠨᠠᠷ (ᠪᠢᠳᠡ᠂ᠲᠡᠳᠡᠨᠡᠷ) ᠶᠠᠪᠤᠭᠤᠯᠵᠠᠭᠠᠴᠢᠬᠠ TANAR/Rb23(BIDE/Rb13TEDENER/Rb33)→YABV/Ve2+GVL/Fe11+JAGA/Fe5+CIH_A/Fi21 The following is A0 Rule Set ᠪᠢ ᠪᠢᠳᠡ ᠴᠢᠲᠠᠨᠠᠷ ᠲᠡᠷᠡᠲᠡᠳᠡᠨᠡᠷ ᠶᠠᠪᠤᠵᠠᠢ BI/Rb11BIDE/Rb13CI/Rb21TANAR/Rb23TERE/Rb31 TEDENER/Rb33→YABV/Ve2+JAI/Fs11 ᠪᠢ ᠪᠢᠳᠡ ᠴᠢᠲᠠᠨᠠᠷ ᠲᠡᠷᠡᠲᠡᠳᠡᠨᠡᠷ (ᠨᠠᠮᠠ ᠪᠢᠳᠡ᠂ᠴᠢᠮᠠ ᠲᠠᠨᠠᠷ ᠲᠡᠭᠦᠨ ᠲᠡᠳᠡᠨᠡᠷ) ᠶᠠᠪᠤᠭᠤᠯᠵᠠᠢ BI/Rb11BIDE/Rb13CI/Rb21TANAR/Rb23TERE/Rb31 TEDENER/Rb33 (NAM_A/Rb12BIDE/Rb13CIM_A/Rb22TANAR/Rb23TEGUN/Rb32TEDENER/Rb33) →YABV/Ve2+GVL/Fe11+JAI/Fs11 ᠪᠢᠳᠡᠲᠠᠨᠠᠷ ᠲᠡᠳᠡᠨᠡᠷ ᠶᠠᠪᠤᠵᠠᠭᠠᠵᠠᠢ BIDE/Rb13TANAR/Rb23TEDENER/Rb33→YABV/Ve2+JAGA/Fe5+JAI/Fs11 ᠪᠢ ᠪᠢᠳᠡ ᠴᠢᠲᠠᠨᠠᠷ ᠲᠡᠷᠡᠲᠡᠳᠡᠨᠡᠷ ᠶᠠᠪᠤᠴᠢᠬᠠᠵᠠᠢ BI/Rb11BIDE/Rb13CI/Rb21TANAR/Rb23TERE/Rb31 TEDENER/Rb33→YABV/Ve2+CIHA/Fi21+JAI/Fs11 ᠪᠢᠳᠡᠲᠠᠨᠠᠷ ᠲᠡᠳᠡᠨᠡᠷ (ᠪᠢᠳᠡ᠂ᠲᠠᠨᠠᠷᠲᠡᠳᠡᠨᠡᠷ) ᠶᠠᠪᠤᠭᠤᠯᠵᠠᠭᠠᠵᠠᠢ BIDE/Rb13TANAR/Rb23TEDENER/Rb33(BIDE/Rb13TANAR/Rb23TEDENER/Rb33)→YABV/Ve2+GVL/Fe11+JAGA/Fe5+JAI/Fs11 ᠪᠢᠳᠡᠲᠠᠨᠠᠷ ᠲᠡᠳᠡᠨᠡᠷ ᠶᠠᠪᠤᠵᠠᠭᠠᠴᠢᠬᠠᠵᠠᠢ BIDE/Rb13TANAR/Rb23TEDENER/Rb33→YABV/Ve2+JAGA/Fe5+CIH_A/Fi21+JAI/Fs11 ᠪᠢᠳᠡᠲᠠᠨᠠᠷ ᠲᠡᠳᠡᠨᠡᠷ (ᠪᠢᠳᠡ᠂ᠲᠠᠨᠠᠷᠲᠡᠳᠡᠨᠡᠷ)ᠶᠠᠪᠤᠭᠤᠯᠵᠠᠭᠠᠴᠢᠬᠠᠵᠠᠢ BIDE/Rb13TANAR/Rb23TEDENER/Rb33(BIDE/Rb13TANAR/Rb23TEDENER/Rb33) →YABV/Ve2+GVL/Fe11+JAGA/Fe5+CIH_A/Fi21+JAI/Fs11 …… Although these rules appear very complicated, there are certain laws andlarge coverage.We provide these rules to the computer through the training set of machine learning and other various channels so as to make up the deficiencies brought about by the "sparsedata"of a small language, to improvethe accuracy of machine learning , and to make the "learning""deeper".
@InProceedings{SUN18.8, author = {Na Sun}, title = {A Word and Its Rules }, booktitle = {Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018)}, year = {2018}, month = {may}, date = {7-12}, location = {Miyazaki, Japan}, editor = {Erhong Yang and Le Sun}, publisher = {European Language Resources Association (ELRA)}, address = {Paris, France}, isbn = {979-10-95546-29-0}, language = {english} }
Summary of the paper