Semantic frame identification with distributed word representations
View Patent ↗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.
1. A computer-implemented method, comprising:
receiving, at a server having one or more processors, labeled training data including a plurality of groups of words, each group of words having a predicate word, each word having generic word embeddings;
extracting, at the server, the plurality of groups of words in a syntactic context of their predicate words;
concatenating, at the server, the generic word embeddings to create a high dimensional vector space representing features for each word;
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; and
outputting, by the server, the model for storage, the model being configured to identify a specific semantic frame for an input.
2. The computer-implemented method of claim 1 , wherein obtaining the model includes training, at the server, the model based on the learned mapping and the learned embeddings.
3. The computer-implemented method of claim 2 , wherein the labeled training data includes (i) frames for verbs and (ii) possible semantic roles for each frame, and wherein modifier roles in the labeled training data are shared across different frames.
4. The computer-implemented method of claim 1 , wherein the learned mapping and the learned embeddings are determined by the server using a linear transformation algorithm.
5. The computer-implemented method of claim 4 , wherein determining the learned mapping and the learned embeddings includes using the linear transformation algorithm with a weighted approximate-rank pairwise loss learned with a stochastic gradient.
6. The computer-implemented method of claim 1 , wherein the features include at least one of (i) direct dependents from a dependency parse tree for a specific predicate word and (ii) dependency paths from the dependency parse tree for each word.
7. The computer-implemented method of claim 1 , further comprising learning, at the server, a technique for filling semantic roles for an identified semantic frame using a rule-based candidate argument extraction algorithm, wherein obtaining the model is further based on the learned technique, and wherein the model is further configured to identify specific roles for a specific semantic frame.
8. The computer-implemented method of claim 1 , further comprising:
indexing, at the server, a plurality of web pages using the model to obtain an indexed plurality of web pages; and
utilizing, by the server, the indexed plurality of web pages to provide search results in response to a search query.
9. The computer-implemented method of claim 1 , further comprising:
receiving, at the server, speech input representing a question;
converting, at the server, the speech input to a text;
analyzing, at the server, the text using the model; and
generating and outputting, by the server, an answer to the question based on the analyzing of the text using the model.
10. The computer-implemented method of claim 1 , further comprising:
receiving, at the server, a text to be translated from a source language to a target language, the source language being a same language as a language associated with the model;
analyzing, at the server, the text using the model; and
generating and outputting, by the server, a translation of the text from the source language to the target language based on the analyzing of the text using the model.
11. A server having one or more processors configured to perform operations comprising:
receiving labeled training data including a plurality of groups of words, each group of words having a predicate word, each word having generic word embeddings;
extracting the plurality of groups of words in a syntactic context of their predicate words;
concatenating the generic word embeddings to create a high dimensional vector space representing features for each word;
obtaining 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; and
outputting the model for storage, the model being configured to identify a specific semantic frame for an input.
12. The server of claim 11 , wherein obtaining the model includes training the model based on the learned mapping and the learned embeddings.
13. The server of claim 12 , wherein the labeled training data includes (i) frames for verbs and (ii) possible semantic roles for each frame, and wherein modifier roles in the labeled training data are shared across different frames.
14. The server of claim 11 , wherein the learned mapping and the learned embeddings are determined by the server using a linear transformation algorithm.
15. The server of claim 14 , wherein determining the learned mapping and the learned embeddings includes using the linear transformation algorithm with a weighted approximate-rank pairwise loss learned with a stochastic gradient.
16. The server of claim 11 , wherein the features include at least one of (i) direct dependents from a dependency parse tree for a specific predicate word and (ii) dependency paths from the dependency parse tree for each word.
17. The server of claim 11 , wherein the operations further comprise learning a technique for filling semantic roles for an identified semantic frame using a rule-based candidate argument extraction algorithm, and wherein obtaining the model is further based on the learned technique, and wherein the model is further configured to identify specific roles for a specific semantic frame.
18. The server of claim 11 , wherein the operations further comprise:
indexing a plurality of web pages using the model to obtain an indexed plurality of web pages; and
utilizing the indexed plurality of web pages to provide search results in response to a search query.
19. The server of claim 11 , wherein the operations further comprise:
receiving speech input representing a question;
converting the speech input to a text;
analyzing the text using the model; and
generating and outputting an answer to the question based on the analyzing of the text using the model.
20. The server of claim 11 , wherein the operations further comprise:
receiving a text to be translated from a source language to a target language, the source language being a same language as a language associated with the model;
analyzing the text using the model; and
generating and outputting a translation of the text from the source language to the target language based on the analyzing of the text using the model.