Phrase-Level Handwriting Prediction Using Segmented N-Gram Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current auto-complete text entry tools are limited to operating at the word level, forcing users to slow their handwriting or text entry rate and interrupt their train of thought to correct predictions, and lack user-provided input for making determinations beyond explicit characters.
Innovation Solution
Implementing a text phrase predictor that accepts handwritten abbreviated text and supplemental hints, parsing these inputs to generate probabilities for full text phrases, sentences, or paragraphs, allowing for display of predicted text based on likelihood.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If predictive auto-complete text entry operates at the word level only, then the system complexity remains low, but the text entry efficiency and accuracy deteriorate due to frequent interruptions and corrections
Solution Approach 1:
The patent segments the text prediction problem into hierarchical levels: character level, word level, and phrase/sentence level. The system processes input progressively through these segments, allowing predictions to operate at multiple granularities simultaneously. This segmentation enables the system to achieve higher productivity by predicting entire phrases while maintaining manageable complexity through modular processing at each level.
Solution Approach 2:
The patent adds a temporal and contextual dimension to traditional word-level prediction by implementing n-gram models that consider sequences of words and phrases. This dimensional expansion allows the system to predict beyond single words to entire phrases and sentences, improving text entry efficiency while the modular architecture manages the increased complexity through structured probabilistic modeling.
2Speed
If users write faster without corrections, then the text entry speed improves, but the accuracy of predictions deteriorates due to reduced time for processing and correction
Solution Approach 1:
The system performs preliminary probabilistic analysis of multiple possible predictions in advance, maintaining a ranked list of candidate phrases and sentences as the user writes. This preliminary action allows the system to prepare multiple prediction options before the user completes their input, enabling fast text entry speed while maintaining high accuracy through pre-computed probabilistic assessments that can be quickly evaluated without interrupting the user's writing flow.
Solution Approach 2:
The system implements continuous feedback by monitoring the user's handwriting in real-time, comparing it against the ranked list of candidate predictions, and dynamically updating the probability assessments. This feedback mechanism allows the system to maintain accuracy even at high writing speeds by continuously refining predictions based on incoming input, with the most likely prediction always available for immediate confirmation or correction.
3Measurement precision
If the system accepts only explicit characters as input, then the input processing is simple, but the prediction accuracy deteriorates due to lack of contextual information
Solution Approach 1:
The patent implements a universal input handling framework that processes multiple types of input signals through a unified probabilistic model. The system can accept explicit characters, implicit contextual cues, temporal patterns, and stylistic features all through the same n-gram based architecture. This multi-functionality improves prediction accuracy by leveraging diverse input information while managing complexity through a single integrated processing framework that handles all input types consistently.
Solution Approach 2:
The system introduces probabilistic n-gram models as intermediaries between raw input and final predictions. These intermediary models transform various input types (explicit characters, contextual patterns, temporal sequences) into standardized probability distributions over candidate phrases and sentences. This intermediary layer improves accuracy by systematically integrating diverse input information while managing complexity through well-defined probabilistic transformations rather than ad-hoc processing for each input type.
Data Source
AI summary
Methods, mobile electronic devices, and computer program products are provided for accepting abbreviated handwritten entry of phrases, sentences or paragraphs, and supplemental information (“hints”), and probabilistically determining the most likely translation of the abbreviated handwriting to a full text counterpart, and displaying same. Abbreviated handwriting is accepted and parsed according to a predefined reduction pattern to produce multi-character handwritten segments and corresponding segment hints. The multi-character handwritten segments and corresponding segment hints are evaluated using n-gram knowledge and/or language models to identify the most likely words corresponding to the multi-character handwritten segments. The most likely words are used to evaluate the abbreviated handwritten entry at the phrase level by evaluating the likelihood of transition from one word to the next amongst the most likely words, to compute phrase probabilities for various combinations of the most likely words. The most likely phrase(s) are output based in part on the phrase probabilities.