Microsoft Technology Licensing, LLC

Patrick Pantel

Michael Gamon

Ahmed Hassan Awadallah

Mark James Encarnacion

Ariel Fuxman

Ashequl Qadir

Madian Khabsa

Ryen William White

Anitha Kannan

Yu Su

Thomas Lin

Li Deng

Jianfeng Gao

Yi-Min Wang

Susan Theresa Dumais

Johnson T Apacible

William B Dolan

Yun-Cheng Ju

Xiaodong He

Jaime Brooks Teevan

Christopher John Brockett

Bernhard Kohlmeier

Victor Poznanski

Bernhard S J Kohlmeier

Pradeep Chilakamarri

Xinying Song

Nikrouz Ghotbi

Nirupama Chandrasekaran

Sujay Kumar Jauhar

Anantha Deepthi Uppala

Vanessa Graham Murdock

Evelyne Viegas

Piali Choudhury

Arjun Mukherjee

Mathieu Etienne Jacques Audouin

Noelle Mallory Sophy

Hassan Sajjad

Svitlana Volkova

Luis Carlos Vargas Herring

Varish Mulwad

Tae Yano

Chu-Cheng Lin

Dongyeop Kang

Woon Kiat Wong

A task agnostic framework for neural model transfer from a first language to a second language, that can minimize computational and monetary costs by accurately forming predictions in a model of the second language by relying on only a labeled data set in the first language, a parallel data set between both languages, a labeled loss function, and an unlabeled loss function. The models may be trained jointly or in a two-stage process.

Minimization of computational demands in model agnostic cross-lingual transfer with neural task representations as weak supervision

A content creation application can include a feature that receives an inline note within a document and communicates the content of the inline note and a user identifier associated with an author of the inline note to an intelligence service. The intelligence service can identify, from the content of the inline note, one or more agents and a request, the identified one or more agents being the author, one or more person agents, one or more bot agents, or a combination thereof. Based on the identified agent (or lack thereof), the intelligence service can generate a message to each of the one or more agents and communicate the message to the each of the one or more agents over a communication channel. A person agent or the author can receive the message and view the message using the appropriate communication application without accessing the original document.

Hybrid intelligence approach to eliciting knowledge for inline notes

This disclosure describes techniques and architectures that involve a latent activity model for workplace emails. Such a model is based, at least in part, on a concept that communications, such as email at a workplace, are purposeful and organized by activities. An activity is a set of interrelated actions and events around a common goal, involving a particular group of people, set of resources, and time framework, for example. The latent activity model involves a probabilistic inference in graphical models that jointly captures the interplay between latent activities and the email contexts governed by the emails. Such contexts may be email recipients, subject and body of the email, and so on.

Activity modeling in email or other forms of communication

A method includes receiving an email addressed to a recipient user, processing the received email using a reparametrized recurrent neural network model to identify an action based on the received email, and wherein the reparametrized recurrent neural network model has been trained on an email dataset annotated with recipient corresponding actions and reparametrized on unannotated conversation data having structures similar to email data.

AI system to determine actionable intent

Subject matter involves using natural language to Web application program interfaces (API), which map natural language commands into API calls, or API commands. This mapping enables an average user with little or no programming expertise to access Web services that use API calls using natural language. An API schema is accessed and using a specialized grammar, with the help of application programmers, canonical commands associated with the API calls are generated. A hierarchical probabilistic distribution may be applied to a semantic mesh associated with the canonical commands to identify elements of the commands that require labeling. The identified elements may be sent to annotators, for labeling with NL phrases. Labeled elements may be applied to the semantic mesh and probabilities, or weights updated. Labeled elements may be mapped to the canonical commands with machine learning to generate a natural language to API interface. Other embodiments are described and claimed.

Natural language interface to web API

Aspects of the technology described herein provide an efficient user interface that enables users to respond to tasks quickly by providing automated quick task notifications via an audio channel. An audio channel quick task system includes components for recognizing and extracting quick tasks from content (e.g., interpersonal communications, composed content, line of business (LOB) application documents), and for prioritizing and routing the quick tasks to the user via an audio channel at an appropriate and relevant time. The system is enabled to process a user response, determine an action for handling the quick task, and execute the action on behalf of the user (e.g., pass a reply to a requestor, pass an instruction to an application or service, queue the quick task notification, delegate the quick task to another user or bot, forward the quick task to a companion device, or launch an application on a companion device).

Automated quick task notifications via an audio channel

In one embodiment, a web service engine servermay predict a successive action by a user based on an entity reference. The web service engine servermay identify an entity referencein a data transmission caused by a user. The web service engine servermay determine from the data transmission a user intention towards the entity referenceusing an intention model based on a transmission log. The web service engine servermay predict a related successive web action optionfor the entity referencebased on the user intention.

Modeling actions for entity-centric search

Various technologies pertaining to exploratory suggestions are described herein. A computer-implemented graph is constructed, where the graph includes nodes that are representative of aspects and edges that are representative of associations between aspects. An aspect is representative of a sub-topic of a topic or a sub-task of a task. The computer-implemented graph is learned based upon content of search logs, and is used to output exploratory suggestions, where a user is exploring a topic or attempting to complete a multi-step task.

Constructing a graph that facilitates provision of exploratory suggestions

An 'Engagement Predictor' provides various techniques for predicting whether things and concepts (i.e., 'nuggets') in content will be engaging or interesting to a user in arbitrary content being consumed by the user. More specifically, the Engagement Predictor provides a notion of interestingness, i.e., an interestingness score, of a nugget on a page that is grounded in observable behavior during content consumption. This interestingness score is determined by evaluating arbitrary documents using a learned transition model. Training of the transition model combines web browsing log data and latent semantic features in training data (i.e., source and destination documents) automatically derived by a Joint Topic Transition (JTT) Model. The interestingness scores are then used for highlighting one or more nuggets, inserting one or more hyperlinks relating to one or more nuggets, importing content relating to one or more nuggets, predicting user clicks, etc.

Predicting interesting things and concepts in content

An 'Interestingness Modeler' uses deep neural networks to learn deep semantic models (DSM) of 'interestingness.' The DSM, consisting of two branches of deep neural networks or their convolutional versions, identifies and predicts target documents that would interest users reading source documents. The learned model observes, identifies, and detects naturally occurring signals of interestingness in click transitions between source and target documents derived from web browser logs. Interestingness is modeled with deep neural networks that map source-target document pairs to feature vectors in a latent space, trained on document transitions in view of a &#x201c;context&#x201d; and optional &#x201c;focus&#x201d; of source and target documents. Network parameters are learned to minimize distances between source documents and their corresponding &#x201c;interesting&#x201d; targets in that space. The resulting interestingness model has applicable uses, including, but not limited to, contextual entity searches, automatic text highlighting, prefetching documents of likely interest, automated content recommendation, automated advertisement placement, etc.

Modeling interestingness with deep neural networks

In one embodiment, a web service engine servermay predict a successive action by a user based on an entity reference. The web service engine serveridentifies an entity referencein a data transmission caused by a user. The web service engine serverdetermines from the data transmission a user intention towards the entity referenceusing an intention model based on a transmission log. The web service engine serverpredicts a related successive web action optionfor the entity referencebased on the user intention.

A search system that automatically generates questions to refine an underspecified query. The system may generate questions even for queries against a database that contains unstructured textual descriptions of items, allowing the system to operate on a database of items that can be constructed inexpensively. The system extracts from the unstructured text combinations of words that may serve as a set of attribute values. The system uses a classifier to filter out attribute values from the set that would generate unanswerable questions. The remaining attribute values are ranked on their ability to narrow the search results and the highest ranking attribute value is used to generate a question to the user who submitted the query. The response to the question narrows the search results, and the process can be repeated iteratively until the search results are sufficiently narrow.

Search system with query refinement

Growing community of inventors

Bellevue, WA, United States of America

Patrick Pantel