Intelligent routing services and systems
A source content routing system is described for distributing source content received from clients such as documents, to translators for performing translation services on the source content. The routing system extracts source content features, which may be represented as vectors. The vectors may be assembled into an input matrix, which may be processed using an artificial neural network, classifier, perceptron, CRF model, and/or the like, to select a translator such as a machine translation system and/or human. The translator provides translation services translation from a source language to a target language, post translation editing, proof reading, quality analysis of a machine, quality analysis of human translation, and/or the like and returns the product to the content routing system or clients.
1. A router for distributing source content from clients to translators for translation services, the router comprising a processor and memory, the processor executing instructions in the memory to:
obtain extracted source content feature vectors from source content feature vectors of source content received from a client;
select a translator using a weight matrix representing connections between layers of a neural network, the neural network configured to receive a target matrix and an input matrix, the input matrix based on the extracted source content feature vectors, and generate a selection of a translator from the target matrix; and
a server for receiving the source content feature vectors from the client, and for transferring the received source content to the selected translator for performing translation services.
2. The router of claim 1 , wherein the processor is configured to assemble the input matrix from the source content feature vectors by using the extracted source content feature vectors as input for columns of the input matrix.
3. The router of claim 1 , wherein the processor is configured to use a summarization module, a keywords and key-phrases module, a domains module, an entities module, a complexity module, and a machine translation suitability module.
4. The router of claim 1 , wherein the router is a component of a server that includes a network interface configured for receiving the source content from the client via a network and routing the source content to the selected translator via the network.
5. The router of claim 1 , wherein the processor is configured to:
receive a translator feature for each of a plurality of translators;
receive a plurality of job features; and
receive the source content for the translation services.
6. The router of claim 5 , wherein the processor is configured to process the input matrix using the target matrix and the weight matrix for the neural network to select the translator from the plurality of translators.
7. The router of claim 6 , wherein the source content feature vectors include a summarization of the source content, a plurality of keywords and key-phrases in the source content, an identification of a domain of the source content, a recognition of a plurality of named entities in the source content, a calculation of complexity of the source content, and a calculation of suitability of the source content for machine translation.
8. The router of claim 7 , wherein the processor is configured to:
generate a translator feature vector for each of the plurality of translators, each translator feature vector including one or more values representing a feature of the translator; and
assemble the target matrix using each translator feature vector as input for columns of the matrix.
9. The router of claim 1 , wherein the source content feature vectors include a plurality of summarization features of a document including selecting sentences from the source content.
10. The router of claim 1 , wherein the processor is configured to:
select a plurality of sentences representing a document summary feature for inclusion in a summary vector, the sentences selected by:
encoding each sentence of the document into a vector representation;
obtaining a ‘centroid’ vector as a sum of these vector representations; and
selecting sentences having a highest cosine similarity to the centroid for inclusion in the summary vector; and
assemble the input matrix using the summary vector as input for a column of the matrix.
11. The router of claim 1 , wherein the processor is configured to:
select a plurality of keywords and key-phrases;
generate a keyword vector representation of the plurality of keywords and key-phrases; and
assemble the input matrix using the keyword vector as input for a column of the matrix.
12. The router of claim 11 , wherein the processor is configured to use at least one of nonparametric spherical topic modeling with word embeddings, non-parametric latent Dirichlet analysis, and Tf/Idf analysis.
13. The router of claim 1 , wherein the processor is configured to:
identify one or more domains using multilayer perceptron with inverse document frequency weighted bag-of-words feature vector;
generate a domain feature vector representation of the one or more identified domains; and
assemble the input matrix using the domain feature vector as input for a column of the matrix.
14. The router of claim 1 , wherein the processor is configured to:
identify one or more named entities using a Conditional Random Field model for entity recognition;
generate an entity feature vector representation of the one or more identified named entities; and
assemble the input matrix using the entity feature vector as input for a column of the matrix.
15. The router of claim 1 , wherein the processor is configured to:
calculate a plurality of complexity values;
generate a complexity feature vector including the plurality of complexity values; and
assemble the input matrix using the complexity feature vector as input for a column of the matrix.
16. The router of claim 15 , wherein the processor is configured to calculate a syntactic complexity, a lexical complexity, an uber index, a Flesch Kincaid score, and an overall complexity.
17. A system for routing source content to translators for translation services, the system comprising:
a router including an artificial neural network and a weight matrix representing connections between layers of the artificial neural network, the router configured to receive a target matrix and an input matrix for a source document; and generate a selection of a translator from the target matrix; and
a server that transfers the source content to the selected translator for performing translation services.
18. The system of claim 17 , wherein the router uses source content feature vectors of the source document as input for columns of the input matrix.
19. The system of claim 18 , wherein the router is configured to
receive a job vector representing a plurality of features and using the job vector as a column of the input matrix.
20. The system of claim 19 , wherein the router is configured to:
receive a plurality of translator feature vectors, each translator feature vector representing a plurality of translation features for a translator of a plurality of translators; and
assemble the target matrix using the translator feature vectors as columns of the target matrix.