Sequence Expander Using Conditional Language Model for Data Entry
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Data entry into electronic devices such as smartphones and smart watches is time-consuming and error-prone, especially for sequences like text, emoji, and pictograms, due to the need for manual input of each indicator, and existing solutions like pre-canned replies and query suggestion systems have limitations in relevance and flexibility.
Innovation Solution
A sequence expander system using a trained conditional language model, integrated with a search component and neural network technology, generates candidate expanded sequences of indicators for efficient data entry and query expansion, allowing users to input subsets of data for full sentences or queries, without relying on computationally expensive n-gram language models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If manual input of each indicator is used for data entry, then data entry can be performed with simple device interface, but data entry becomes time-consuming and error-prone
Solution Approach 1:
The system performs preliminary actions by predicting and generating candidate expanded sequences before the user completes input. The language model pre-computes multiple possible completions based on the partial input, allowing the user to simply select from pre-generated options rather than typing everything manually.
Solution Approach 2:
The patent introduces an intermediary component - the language model with candidate sequence generator - that mediates between the user's partial input and the final complete sequence. This intermediary automatically expands the input by generating and ranking candidate sequences, reducing the user's manual input burden while maintaining accuracy.
2Productivity
If pre-canned standard replies are used for messaging, then response time can be reduced, but reply flexibility and relevance are limited
Solution Approach 1:
The system transitions from static pre-canned replies to dynamic generated sequences. The language model adapts its output based on the specific context, conversation history, and user input, generating customized replies in real-time rather than selecting from fixed templates. This makes the reply system as flexible as manual typing but as fast as template selection.
Solution Approach 2:
The system changes the parameter of reply generation from fixed predefined options to dynamically generated variations. By adjusting the language model's generation parameters and using context-aware processing, the system produces replies that are both quick to obtain and highly adaptable to specific conversational contexts.
3Measurement precision
If query expansion is performed using traditional methods, then search relevance can be improved, but computational complexity and processing time increase
Solution Approach 1:
The system performs partial expansion by generating candidate sequences that extend beyond the minimal necessary expansion. The language model generates multiple candidate completions with varying degrees of expansion, and the system selects the most relevant ones, achieving good search relevance without exhaustively exploring all possible expansions.
Solution Approach 2:
The patent uses computationally inexpensive language model generations as disposable candidates. Multiple candidate sequences are generated quickly and cheaply, then filtered and ranked based on relevance. This approach trades the generation of many low-cost candidate sequences for avoiding the need for complex, computationally expensive traditional query expansion methods.
4Productivity
If next-word prediction is used in virtual keyboards, then typing speed can be improved, but prediction accuracy and correction capability are limited
Solution Approach 1:
The system segments the prediction task into generating multiple independent candidate sequences rather than predicting a single next word. By dividing the prediction into multiple parallel candidate generations, each candidate can be optimized for different aspects of accuracy, and the user can select the most accurate completion.
Solution Approach 2:
The language model serves multiple functions: it acts as a next-word predictor, a sentence completion generator, and a context-aware suggestion system. This multi-functional approach allows the system to handle various prediction scenarios with a single unified model, improving overall prediction accuracy across different contexts and input types.
Data Source
AI summary
An electronic device is described which has a user interface which receives an input comprising a sequence of target indicators of data items. The data entry system has a search component which searches for candidate expanded sequences of indicators comprising the target indicators. The search component searches amongst indicators generated by a trained conditional language model, the conditional language model having been trained using pairs, each individual pair comprising a sequence of indicators and a corresponding expanded sequence of indicators.