Semantic Frame Identification via Distributed Word Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Natural language parsing, specifically semantic frame identification, is challenging due to the ambiguity of grammar in natural languages, which leads to multiple possible parses for a sentence, and existing methods struggle to accurately represent word meanings in context using distributed word representations.
Innovation Solution
A computer-implemented technique that receives labeled training data to extract and concatenate word embeddings, creating a high-dimensional vector space, and then maps this to a low-dimensional space using a linear transformation algorithm to identify specific semantic frames, enabling improved semantic frame identification and argument filling through rule-based candidate argument extraction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If generic word embeddings are used to represent words, then word meaning representation is simplified, but semantic frame identification accuracy deteriorates due to inability to capture syntactic context
Solution Approach 1:
The patent segments the word representation into two parts: generic word embedding (capturing word meaning) and syntactic context embedding (capturing grammatical role). This segmentation allows each component to specialize in one aspect, resolving the contradiction between simplicity and accuracy.
Solution Approach 2:
The patent merges generic word embeddings with syntactic context embeddings to create a composite representation. This combination integrates the simplicity of word-level representations with the accuracy of context-aware representations, achieving both goals simultaneously.
2Measurement precision
If syntactic context is added to word representations, then semantic frame identification accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent segments the computational task into independent modules: word embedding extraction, syntactic context extraction, and combination. This segmentation allows each module to be optimized independently, reducing overall computational complexity while maintaining accuracy.
Solution Approach 2:
The patent introduces an intermediary representation layer that combines word embeddings and syntactic context embeddings. This intermediary layer acts as a mediator, integrating information efficiently without requiring complex interactions between all input elements, thus reducing computational burden.
3Productivity
If distributed word representations are used, then processing efficiency is improved, but ability to capture nuanced word meanings in context deteriorates
Solution Approach 1:
The patent segments the meaning representation into lexical meaning (from word embeddings) and contextual meaning (from syntactic context). This segmentation preserves the efficiency of distributed representations while adding contextual nuance through a separate, specialized component.
Solution Approach 2:
The patent creates a composite representation by combining multiple types of embeddings (word embeddings, syntactic context embeddings, and potentially semantic role embeddings). This composite approach leverages the strengths of each component: efficiency from distributed representations and accuracy from context-aware representations.
Data Source
AI summary
A computer-implemented technique can include receiving, at a server, labeled training data including a plurality of groups of words, each group of words having a predicate word, each word having generic word embeddings. The technique can include extracting, at the server, the plurality of groups of words in a syntactic context of their predicate words. The technique can include concatenating, at the server, the generic word embeddings to create a high dimensional vector space representing features for each word. The technique can include obtaining, at the server, a model having a learned mapping from the high dimensional vector space to a low dimensional vector space and learned embeddings for each possible semantic frame in the low dimensional vector space. The technique can also include outputting, by the server, the model for storage, the model being configured to identify a specific semantic frame for an input.