Microsoft Technology Licensing, LLC

Mmodal IP LLC

Multimodal Technologies, LLC

Microsfot Technology Licensing, LLC

Anoop Deoras

Ruhi Sarikaya

Fethiye Asli Celikyilmaz

Detlef Koll

Juergen Fritsch

Daniel Boies

Dong Yu

Li Deng

Dilek Zeynep Hakkani-Tur

Gokhan Tur

Xiaodong He

Geoffrey G Zweig

Fileno A Alleva

Mei-Yuh Hwang

Kaisheng N Yao

Zhaleh Feizollahi

Dustin Hillard

Gregoire Mesnil

Ravikiran Janardhana

Technologies pertaining to slot filling are described herein. A deep neural network, a recurrent neural network, and/or a spatio-temporally deep neural network are configured to assign labels to words in a word sequence set forth in natural language. At least one label is a semantic label that is assigned to at least one word in the word sequence.

Assignment of semantic labels to a sequence of words using neural network architectures

An understanding model is trained to account for human perception of the perceived relative importance of different tagged items (e.g. slot/intent/domain). Instead of treating each tagged item as equally important, human perception is used to adjust the training of the understanding model by associating a perceived weight with each of the different predicted items. The relative perceptual importance of the different items may be modeled using different methods (e.g. as a simple weight vector, a model trained using features (lexical, knowledge, slot type, . . . ), and the like). The perceptual weight vector and/or or model are incorporated into the understanding model training process where items that are perceptually more important are weighted more heavily as compared to the items that are determined by human perception as less important.

Using human perception in building language understanding models

Non-verbalized tokens, such as punctuation, are automatically predicted and inserted into a transcription of speech in which the tokens were not explicitly verbalized. Token prediction may be integrated with speech decoding, rather than performed as a post-process to speech decoding.

Decoding-time prediction of non-verbalized tokens

A scalable statistical language understanding (SLU) system uses a fixed number of understanding models that scale across domains and intents (i.e. single vs. multiple intents per utterance). For each domain added to the SLU system, the fixed number of existing models is updated to reflect the newly added domain. Information that is already included in the existing models and the corresponding training data may be re-used. The fixed models may include a domain detector model, an intent action detector model, an intent object detector model and a slot/entity tagging model. A domain detector identifies different domains identified within an utterance. All/portion of the detected domains are used to determine associated intent actions. For each determined intent action, one or more intent objects are identified. Slot/entity tagging is performed using the determined domains, intent actions, and intent object detector.

Scaling statistical language understanding systems across domains and intents

Processes capable of accepting linguistic input in one or more languages are generated by re-using existing linguistic components associated with a different anchor language, together with machine translation components that translate between the anchor language and the one or more languages. Linguistic input is directed to machine translation components that translate such input from its language into the anchor language. Those existing linguistic components are then utilized to initiate responsive processing and generate output. Optionally, the output is directed through the machine translation components. A language identifier can initially receive linguistic input and identify the language within which such linguistic input is provided to select an appropriate machine translation component. A hybrid process, comprising machine translation components and linguistic components associated with the anchor language, can also serve as an initiating construct from which a single language process is created over time.

Building multi-language processes from existing single-language processes

Growing community of inventors

San Jose, CA, United States of America

Anoop Deoras