Abstract:
본 발명에 따른 자동 번역 시스템은 사용자로부터 상기 번역 원문을 입력받고, 상기 번역 사전을 이용하여 자동 번역하는 번역 엔진과, 상기 번역 엔진에서 번역된 결과에 대해 상기 사용자로부터 수정된 부분을 입력받는 번역 지식 편집기와, 상기 번역 지식 편집기에서 수정된 부분이 상기 번역 엔진 상의 오류인지 분석하는 번역 프로파일 학습기와, 상기 번역 프로파일 학습기에서 분석된 결과를 전달받아 상기 번역 사전에 반영하는 번역 사전 반영기를 포함하는 것을 특징으로 한다.
Abstract:
정답 집합의 정답문을 확장하여 정답 집합의 정답문 개수가 서로 다를 때에도 번역 품질을 정확히 측정하도록 한 자동 번역기의 번역 품질 측정을 위한 정답 집합 확장 장치 및 방법이 제시된다. 제시된 자동 번역기의 번역 품질 측정을 위한 정답 집합 확장 장치는 복수의 정답 집합들, 확장 사전, 언어 모델, 확장 정답 집합을 저장하는 저장부; 저장부에 저장된 복수의 정답 집합들 및 확장 사전을 근거로 번역 원문에 대한 하나 이상의 정답문을 생성하는 레퍼런스 확장부; 및 정답문들에 대한 사용자 단말의 승인 여부를 근거로 인증에 성공한 정답문들을 이용하여 확장 정답 집합을 생성하고, 생성한 확장 정답 집합을 저장부에 저장하는 레퍼런스 확장 인증부를 포함한다.
Abstract:
The present invention relates to an apparatus for driving an agent system based on a natural language processing operation. The apparatus for driving the agent system detects next intention of a user by using information contained in a conversation content of the user stored during the operation of an automatic translation system installed in a portable terminal, and copes with the detected intention of the user. Therefore, according to the present invention, input and output sentences of the automatic translation system are utilized as context information, so that the convenience in use of the agent system can be improved.
Abstract:
The present invention relates to an apparatus and method for improving Chinese word segmentation performance, and more particularly, an apparatus and method for improving word segmentation performance by processing word segmentation errors of Chinese by automatically recognizing an accurate boundary of a word from a sentence of another language, for example, English or Korean, of a parallel corpus of which a word boundary is clear in order to reduce unregistered word errors and ambiguity errors frequently appeared in a Chinese word segmenting device. According to the present invention, a limitation that errors are confirmed from the word segmenting device by consuming lots of manpower and time can be overcome by continuously extracting the unregistered word errors and ambiguity errors, which are difficult to process at the time of word segmentation of a Chinese sentence, through the parallel corpus and storing corrected word segmentation information.
Abstract:
According to one embodiment, an automatic learning-based artificial intelligent dialog system includes: a database in which personalized expression learning data including sentence expressions classified by personal intention and classification tags for the sentence expressions is stored; a learning device which analyzes the sentence expressions included in the personalized expression learning data at a morpheme level and learns personal profiling data expressions attached with the classification tags at a morpheme level; a language analysis unit which analyzes the currently-inputted dialog sentence at a morpheme level; an extraction unit which extracts user profiling data based on the analysis result at a morpheme level by the language analysis unit and the personal profiling data expressions; a personal history database in which the user profiling data is classified by personal preference and accumulated as personal profiling data; an intention analysis unit which determines the intention in the dialog sentence based on the analysis result at a morpheme level by the language analysis unit; and a response generation unit which determines a dialog flow of the dialog sentence based on the personal profiling data accumulated in the personal history database and generates a response sentence. [Reference numerals] (10) Database; (20) Learning unit; (30) Language analysis unit; (40) Extraction unit; (50) Database; (60) Intention analysis unit; (70) Response generation unit
Abstract:
PURPOSE: A Korean morpheme analysis device and a method thereof are provided to gradually improve a morpheme analysis performance by providing a self-learning function for analyzing sentences. CONSTITUTION: A morpheme analysis unit (12) analyzes morpheme for a Korean sentence. A correction data extraction unit (144) extracts analysis error correction data by using a sentence clue in the analysis result of the morpheme analysis unit. A storage unit (16) stores the analysis error correction data, and a correction unit (18) outputs the final morpheme analysis result by applying the analysis error correction data to the analysis result of the morpheme analysis unit. [Reference numerals] (10) Input unit; (12) Morpheme analysis unit; (14) Corrected data extraction unit; (16) Storage unit; (18) Correction unit; (AA) Final morpheme analysis result
Abstract:
PURPOSE: A compression and decompression method in a dictionary data structure implementation unit and an apparatus thereof are provided to prepare a separate compression and decompression space with only the small amount of main memories corresponding to a single KVP(Key-Value Pair) in a dictionary when referring to a compressed dictionary file, and not to compress and decompress the part actually not used for the compression and decompression, thereby providing advantages in a speed and a space. CONSTITUTION: A compression device calculates the appearance frequency of the longest matching word in a specific file where data to be used is stored(S310). The compression device arranges the data of the specific file based on the calculated appearance frequency and the length of the corresponding data, and performs encoding by using a frequency table(S320). The compression device generates an encoding table(S330). The compression device compresses original data in a dictionary by using the encoding table(S340). [Reference numerals] (AA) Start; (BB) End; (S310) Extract an appearance frequency; (S320) Ranking and encoding; (S330) Store an encoding table; (S340) Compress original data in a dictionary