Probabilistic Contextual Model for Text Input Term Boundary Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing text input systems, such as those on Android and iOS devices, are limited in detecting term boundaries, particularly when words are mistyped, misspelled, or when automatic apostrophe insertion is involved, and can only analyze single word boundaries, hindering the ability to enter phrases or messages without explicit insertion of term boundaries.
Innovation Solution
A system and method that uses a probabilistic contextual model to infer likely term boundaries in text input by generating candidates from input sequences, assigning probability estimates based on a context language model, and discarding candidates with low probability ratios, allowing for multi-term predictions and correcting mistyping or misspelling while inferring word boundaries and prefixes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing text input systems (Android ICS keyboard or iOS keyboard) are used to detect term boundaries, then single word boundary detection is achieved, but the system cannot handle multiple term boundaries, mistyped words, or automatic apostrophe insertion
Solution Approach 1:
The input sequence is divided into multiple candidates, each representing a different segmentation of term boundaries. The system generates multiple possible interpretations of where term boundaries should be inserted, allowing it to evaluate different segmentations and select the most probable one, thereby resolving the contradiction between detection precision and adaptability to various input scenarios
Solution Approach 2:
The system changes the parameter of term boundary insertion by introducing optional term boundary nodes in the probabilistic constrained sequence graph. This allows the system to dynamically insert or omit term boundaries based on probability calculations, enabling it to handle mistyped words, misspellings, and automatic apostrophe insertion while maintaining accurate term boundary detection
2Measurement precision
If users manually insert term boundaries (spaces) between words, then accurate term separation is achieved, but text entry efficiency decreases due to additional manual input
Solution Approach 1:
The system performs self-service by automatically inferring and inserting term boundaries without requiring manual user input. The probabilistic contextual model autonomously analyzes the input sequence, generates candidates with different term boundary configurations, and selects the most probable segmentation, thereby maintaining accurate term separation while significantly improving text entry efficiency
Solution Approach 2:
The system uses feedback from the probabilistic contextual model to continuously refine term boundary detection. By calculating probabilities based on context language models and discarding low-probability candidates, the system provides feedback-driven automatic term separation that achieves high accuracy without manual intervention, thus resolving the contradiction between precision and productivity
3Device complexity
If the system analyzes only single word boundaries, then computational complexity is reduced, but the ability to predict entire phrases and messages is limited
Solution Approach 1:
The system performs preliminary action by generating multiple candidates with different term boundary configurations before final selection. This advance preparation of multiple possible segmentations allows the system to handle phrase and message prediction comprehensively, and the probabilistic filtering in subsequent steps manages the computational complexity of this extended analysis
Data Source
Figure 1
Figure 2a~2b
Figure 3
AI summary
A system for inputting text into an electronic device. The system comprises a candidate generator(2)configured to generate one or more candidates from an input sequence(20). The input sequence (20) comprises a contiguous sequence of characters. Each candidate comprises two or more terms separated by one or more term boundaries. The candidate generator (2) is configured to assign a first probability estimate to each candidate by: searching for one or more terms of that candidate in a context language model, wherein the context language model comprises sequences of terms, each sequence of terms having a corresponding probability of occurring;and assigning the probability corresponding to the one or more terms of that candidate from the context language model to the candidate. The candidate generator (2) is configured to discard one or more candidates on the basis of the corresponding first probability estimates. There is provided a corresponding method of inferring term boundaries in a user input sequence.