Utterance Sentence Expansion Using Noun Generalization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Utterance generation models struggle to generate appropriate output utterances for unknown inputs that do not include token sequences similar to the training data, particularly when encountering unknown nouns not present in the training data.
Innovation Solution
An utterance sentence expansion device that uses an expansion dictionary to insert higher-level categories of nouns into the input utterance, generating an expanded utterance which is then used to train and improve the output of the utterance generation model, allowing it to handle unknown inputs by relating them to known concepts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If training data is limited to available utterance-token pairs, then the training process is simple and fast, but the model cannot generate appropriate output utterances for unknown inputs that do not include token sequences similar to the training data
Solution Approach 1:
The system performs preliminary action by expanding nouns in training data to higher-level categories before training the model. This preprocessing step ensures that the model learns generalized representations that can handle unknown inputs during inference, without requiring additional training data to be collected later.
Solution Approach 2:
The expansion dictionary serves as a universal resource that maps specific nouns to higher-level categories. This universal mapping mechanism allows the model to generalize from known concepts to unknown concepts across different domains, enhancing adaptability without domain-specific training data.
2Reliability
If the model is trained only on exact token sequences from training data, then training is efficient, but the model fails to generate appropriate responses for utterances containing unknown nouns not present in training data
Solution Approach 1:
The system performs noun expansion to higher-level categories as a preliminary processing step before training. This ensures that the training data contains generalized concepts that improve the reliability of output utterances for unknown inputs, while the expansion dictionary pre-computes the mappings to avoid complexity during training execution.
Solution Approach 2:
The expansion dictionary acts as an intermediary between specific nouns and higher-level categories. It mediates the transformation of concrete nouns into abstract categories during training data preparation, enabling the model to learn generalized patterns without directly handling the complexity of noun hierarchy relationships.
3Adaptability or versatility
If higher-level categories are inserted into utterances using an expansion dictionary, then the model can generalize better to unknown inputs, but the utterance processing becomes more complex
Solution Approach 1:
The noun expansion to higher-level categories is performed as a preliminary processing step on training data before model training. This upfront preparation enhances generalization capability while the expansion dictionary stores pre-computed mappings, avoiding the need to perform complex expansion operations during each training iteration or inference step.
Solution Approach 2:
The expansion dictionary serves itself by storing pre-computed mappings from nouns to higher-level categories. Once the dictionary is built, it automatically provides the necessary expansions without requiring complex processing logic during training or inference, making the system self-sufficient for handling noun generalization.
Data Source
AI summary
An expanded utterance that is used to output a more appropriate output utterance for an utterance can be generated.An utterance sentence expansion device includes an expansion unit that inserts, for an utterance that is an utterance to be expanded that includes a noun and is morphologically analyzed in advance, by using information of an expansion dictionary, which includes higher-level categories of the noun, one or more higher-level categories of the expansion dictionary corresponding to the noun included in the utterance into a position before the noun of the utterance to generate an expanded utterance.