Abstract: Techniques disclosed herein relate generally to language detection. In one particular aspect, a method is provided that includes obtaining a sequence of n-grams of a textual unit; using an embedding layer to obtain an ordered plurality of embedding vectors for the sequence of n-grams; using a deep network to obtain an encoded vector that is based on the ordered plurality of embedding vectors; and using a classifier to obtain a language prediction for the textual unit that is based on the encoded vector. The deep network includes an attention mechanism, and using the embedding layer to obtain the ordered plurality of embedding vectors comprises, for each n-gram in the sequence of n-grams: obtaining hash values for the n-gram; based on the hash values, selecting component vectors from among the plurality of component vectors; and obtaining an embedding vector for the n-gram that is based on the component vectors.

Abstract: Techniques are provided for customizing or fine-tuning a pre-trained version of a machine-learning model that includes multiple layers and is configured to process audio or textual language input. Each of the multiple layers is configured with a plurality of layer-specific pre-trained parameter values corresponding to a plurality of parameters, and each of the multiple layers is configured to implement multi-head attention. An incomplete subset of the multiple layers is identified for which corresponding layer-specific pre-trained parameter values are to be fine-tuned using a client data set. The machine-learning model is fine-tuned using the client data set to generate an updated version of the machine-learning model, where the layer-specific pre-trained parameter values configured for each layer of one of more of the multiple layers not included in the incomplete subset are frozen during the fine-tuning. Use of the updated version of the machine-learning model is facilitated.

Abstract: Techniques are disclosed herein for focused training of language models and end-to-end hypertuning of the framework. In one aspect, a method is provided that includes obtaining a machine learning model pre-trained for language modeling, and post-training the machine learning model for various tasks to generate a focused machine learning model. The post-training includes: (i) training the machine learning model on an unlabeled set of training data pertaining to a task that the machine learning model was pre-trained for as part of the language modeling, and the unlabeled set of training data is obtained with respect to a target domain, a target task, or a target language, and (ii) training the machine learning model on a labeled set of training data that pertains to another task that is an auxiliary task related to a downstream task to be performed using the machine learning model or output from the machine learning model.

Abstract: Techniques disclosed herein relate generally to text classification and include techniques for fusing word embeddings with word scores for text classification. In one particular aspect, a method for text classification is provided that includes obtaining an embedding vector for a textual unit, based on a plurality of word embedding vectors and a plurality of word scores. The plurality of word embedding vectors includes a corresponding word embedding vector for each of a plurality of words of the textual unit, and the plurality of word scores includes a corresponding word score for each of the plurality of words of the textual unit. The method also includes passing the embedding vector for the textual unit through at least one feed-forward layer to obtain a final layer output, and performing a classification on the final layer output.

Abstract: Techniques are provided for using context tags in named-entity recognition (NER) models. In one particular aspect, a method is provided that includes receiving an utterance, generating embeddings for words of the utterance, generating a regular expression and gazetteer feature vector for the utterance, generating a context tag distribution feature vector for the utterance, concatenating or interpolating the embeddings with the regular expression and gazetteer feature vector and the context tag distribution feature vector to generate a set of feature vectors, generating an encoded form of the utterance based on the set of feature vectors, generating log-probabilities based on the encoded form of the utterance, and identifying one or more constraints for the utterance.

Abstract: Techniques are disclosed for systems including techniques for multi-feature balancing for natural langue processors. In an embodiment, a method includes receiving a natural language query to be processed by a machine learning model, the machine learning model utilizing a dataset of natural language phrases for processing natural language queries, determining, based on the machine learning model and the natural language query, a feature dropout value, generating, and based on the natural language query, one or more contextual features and one or more expressional features that may be input to the machine learning model, modifying at least one or the one or more contextual features and the one or more expressional features based on the feature dropout value to generate a set of input features for the machine learning model, and processing the set of input features to cause generating an output dataset for corresponding to the natural language query.

Abstract: Disclosed herein are techniques for addressing an overconfidence problem associated with machine learning models in chatbot systems. For each layer of a plurality of layers of a machine learning model, a distribution of confidence scores is generated for a plurality of predictions with respect to an input utterance. A prediction is determined for each layer of the machine learning model based on the distribution of confidence scores generated for the layer. Based on the predictions, an overall prediction of the machine learning model is determined. A subset of the plurality of layers are iteratively processed to identify a layer whose assigned prediction satisfies a criterion. A confidence score associated with the assigned prediction of the layer of the machine learning model is assigned as an overall confidence score to be associated with the overall prediction of the machine learning model.

Abstract: Techniques for using logit values for classifying utterances and messages input to chatbot systems in natural language processing. A method can include a chatbot system receiving an utterance generated by a user interacting with the chatbot system. The chatbot system can input the utterance into a machine-learning model including a set of binary classifiers. Each binary classifier of the set of binary classifiers can be associated with a modified logit function. The method can also include the machine-learning model using the modified logit function to generate a set of distance-based logit values for the utterance. The method can also include the machine-learning model applying an enhanced activation function to the set of distance-based logit values to generate a predicted output. The method can also include the chatbot system classifying, based on the predicted output, the utterance as being associated with the particular class.

Abstract: Techniques for using enhanced logit values for classifying utterances and messages input to chatbot systems in natural language processing. A method can include a chatbot system receiving an utterance generated by a user interacting with the chatbot system and inputting the utterance into a machine-learning model including a series of network layers. A final network layer of the series of network layers can include a logit function. The machine-learning model can map a first probability for a resolvable class to a first logit value using the logit function. The machine-learning model can map a second probability for a unresolvable class to an enhanced logit value. The method can also include the chatbot system classifying the utterance as the resolvable class or the unresolvable class based on the first logit value and the enhanced logit value.

Abstract: Techniques are disclosed for tuning hyperparameters of a model. Datasets are obtained for training the model and metrics are selected for evaluating performance of the model. Each metric is assigned a weight specifying an importance to the performance of the model. A function is created that measures performance based on the weighted metrics. Hyperparameters are tuned to optimize the model performance. Tuning the hyperparameters includes: (i) training the model that is configured based on a current values for the hyperparameters; (ii) evaluating a performance of the model using the function; (iii) determining whether the model is optimized for the metrics; (iv) in response to the model not being optimized, searching for a new values for the hyperparameters, reconfiguring the model with the new values, and repeating steps (i)-(iii) using the reconfigured model; and (v) in response to the model being optimized for the metrics, providing a trained model.

Abstract: The present disclosure relates to techniques for identifying out-of-domain utterances.

Abstract: Techniques are disclosed for tuning hyperparameters of a machine-learning model. A plurality of metrics are selected for which hyperparameters of the machine-learning model are to be tuned. Each metric is associated with a plurality of specification parameters including a target score, a penalty factor, and a bonus factor. The plurality of specification parameters are configured for each metric in accordance with a first criterion. The machine-learning model is evaluated using one or more validation datasets to obtain a metric score. A weighted loss function is formulated based on a difference between the metric score and the target score of each metric, the penalty factor or the bonus factor. The hyperparameters associated with the machine-learning model are tuned in order to optimize the weighted loss function. In response to the weighted loss function being optimized, the machine-learning model is provided as a validated machine-learning model.