Character Gender Identification with Speaker-Aware Self-Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing character gender identification methods rely on small-scale annotated datasets and external knowledge, leading to inefficiencies and error propagation, and existing methods for inferring gender from names are not reliable or accessible for new languages.
Innovation Solution
A method involving a multi-task self-training paradigm that iteratively trains a model for both speaker identification and gender identification using unlabeled corpora, leveraging large-scale unlabeled book corpora and a new annotation guideline to speed up human annotation and improve performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual annotation of character gender is performed for each book, then annotation quality is improved, but annotation cost and time consumption increase significantly
Solution Approach 1:
The patent applies preliminary action by pre-training a multi-task model on large-scale unlabeled book corpora before performing gender identification. The model is iteratively trained to perform both speaker identification and gender identification, preparing the system in advance to handle new books efficiently without requiring manual annotation of each book from scratch.
Solution Approach 2:
The patent implements self-service through the self-training mechanism where the model uses its own predictions on unlabeled data to improve its performance. The system automatically generates training data by predicting gender labels on unlabeled book corpora, allowing the model to train itself without human annotators for each individual book.
2Productivity
If large-scale annotated datasets are constructed, then model training capability is improved, but data collection cost and complexity increase
Solution Approach 1:
The patent uses copying by creating synthetic training data through the multi-task self-training process. Instead of collecting real annotated data from books, the system generates pseudo-labeled training instances by having the model predict gender labels on large-scale unlabeled book corpora, effectively copying the annotation task through automated prediction.
Solution Approach 2:
The patent introduces an intermediary mechanism in the form of the multi-task model that bridges the gap between unlabeled data and training requirements. The model serves as a mediator that processes unlabeled book corpora and transforms them into useful training instances for both speaker identification and gender identification tasks.
3Productivity
If existing NER and coreference resolution methods are used for gender identification, then processing efficiency is improved, but accuracy deteriorates due to error propagation
Solution Approach 1:
The patent merges speaker identification and gender identification into a single multi-task model. Instead of using separate NER and coreference resolution pipelines that can propagate errors, the integrated model performs both tasks simultaneously, allowing the gender identification to benefit from the speaker identification context and vice versa, thereby improving overall reliability.
Data Source
AI summary
A method and apparatus that identifies one or more characters within a text; determines one or more informative sections within the text, the one or more informative sections providing information regarding a gender of the one or more characters within the text; selects a most informative section from the one or more informative sections; extracts unlabeled instances corresponding to the gender of the one or more characters from the most informative section; iteratively trains a multi-task model using unlabeled corpora, the multi-task model performing both speaker identification and gender identification; and labels the gender of the one or more characters based on the extracted unlabeled instances and the multi-task model.