Predicting Next Syllable in Agglutinative Language Input
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In agglutinative languages like Korean, Japanese, and Turkish, the large number of words derived using affixes makes it difficult to create an effective language model for recommending the next syllable, as existing methods require extensive data and struggle to accurately predict words not registered in the dictionary.
Innovation Solution
A device with a memory storing sequences of registered words, an input unit to receive text, and a controller using neural network algorithms to detect registered words and predict the next syllable by applying filters to syllables and morphemes, even if the input syllable is not registered, thereby refining the language model to suggest probable next syllables.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a large number of words are registered in the language model to cover all possible affixed words in agglutinative languages, then the coverage of the language model is improved, but the data requirements and system complexity increase significantly
Solution Approach 1:
The patent segments words into root words and affixes separately. Instead of registering every possible affixed word combination, the system stores root words and their associated affixes independently. When processing input, it identifies the root word and matches applicable affixes from the stored set, thereby covering all possible combinations without requiring explicit registration of each full word form.
Solution Approach 2:
The patent creates a universal language model structure that handles multiple word forms through a single root word entry. Each root word in the language model serves multiple functions by combining with different affixes from a shared affix set, allowing the same root word data structure to generate predictions for numerous derived words without requiring separate entries for each.
2Measurement precision
If all possible word combinations with affixes are registered in the language model, then prediction accuracy for any word form is improved, but the memory and processing requirements become unmanageable
Solution Approach 1:
The patent divides the language model into two separate components: a root word dictionary and an affix set. This segmentation allows the system to store only essential linguistic units (roots and affixes) rather than every possible word combination. The memory requirement grows linearly with the number of roots and affixes rather than exponentially with their combinations.
Solution Approach 2:
The patent implements a nested structure where affixes are associated with root words in a hierarchical manner. Each root word contains a set of applicable affixes, forming a nested organization. This nesting allows the system to efficiently access and combine root words with their relevant affixes without requiring flat storage of all possible combinations.
3Reliability
If the language model includes every possible affixed word variation, then the ability to recommend exact words is improved, but the system becomes difficult to maintain and update
Solution Approach 1:
The patent segments the vocabulary into maintainable components: root words and affixes. When new words need to be added or existing ones updated, the system only requires modifying the relevant root word entry or affix set rather than updating numerous pre-combined word forms. This segmentation dramatically reduces the effort and error-proneness of maintaining the language model.
4Adaptability or versatility
If extensive data is collected to cover all word forms in agglutinative languages, then the language model's comprehensiveness is improved, but the processing time and computational resources increase
Solution Approach 1:
The patent extracts and separates the affix components from complete word forms. Instead of processing entire affixed words, the system extracts the root word and applicable affixes independently. This extraction reduces the processing burden by focusing computation on identifying the root and selecting from a predefined affix set, rather than searching through all possible word combinations.
Solution Approach 2:
The patent performs preliminary organization of affixes during system setup, creating ready-to-use affix sets associated with each root word. This preliminary action allows the runtime processing to simply match input against pre-organized structures rather than performing complex computations to generate all possible word forms, significantly improving processing speed.
Data Source
AI summary
Provided is a device including a memory storing information about sequences of a plurality of registered words; an input unit comprising input circuitry configured to receive an input of a text comprising a first eojeol not belonging to the plurality of registered words, wherein, in the first eojeol, a first word is attached to a first registered word that belongs to the plurality of registered words; and a controller configured to detect the first registered word from the first eojeol, to determine a predicted eojeol to be input after the text, based on the information about the sequences of the plurality of registered words and the detected first registered word and to control a display to display the predicted eojeol.