International Business Machines Corporation

Oath Inc.

University of Michigan

Dragomir Radkov Radev

Eric William Brown

Amanda Joy Stent

John Martin Prager

Anni Rosa Coden

Yashar Mehdad

Kapil Raj Thadani

Youssef Billawala

Valerie Samn

David Nahamoo

Rui Zhang

Wlodek Wlodzimierz Zadrozny

Nandakishore Kambhatla

Lazaros Polymenakos

Yiming Ye

Tetsunosuke Fujisaki

Honglak Lee

Catherine Gody Wolf

Youssef Billawal

Joyce Yue Chai

Sunil Subramanyam Govindappa

Systems, methods, and apparatuses are disclosed for adaptively generating a summary of web-based content based on an attribute of a mobile communication device having transmitted a request for the web-based content. By adaptively generating the summary based on an attribute of the mobile communication device such as an amount of visual space available or a number of characters permitted in the interface, a display of the web-based content may be controlled on the mobile communication device in a way that was not previously available. This enables control of displaying web-based content that has been adaptively generated to be displayed on limited display screens based on a learned attribute of the mobile communication device requesting the web-based content.

Scalable and effective document summarization framework

An information processing system, a computer program product, and methods for modeling multi-party dialog interactions. A method includes learning, directly from data obtained from a multi-party conversational channel, to identify particular multi-party dialog threads as well as participants in one or more conversations. Each participant utterance being converted to a continuous vector representation updated in a model of the multi-party dialog relative to each participant utterance and according to each participant's role selected from the set of: sender, addressee, or observer. The method trains the model to choose a correct addressee and a correct response for each participant utterance, using a joint selection criterion. The method learns directly from the data obtained from the multi-party conversational channel, which dialog turns belong to each particular multi-party dialog thread.

Modeling multiparty conversation dynamics: speaker, response, addressee selection using a novel deep learning approach

Automated question answering is disclosed that relates to the selection of an answer to a question from a pool of potential answers which are manually or automatically extracted from a large collection of textual documents. The a feature extraction component, a feature combination component, an answer selection component, and an answer presentation component, among others, are included. The input to the system is a set of one or more natural language questions and a collection of textual document. The output is a (possibly ranked) set of factual answers to the questions, these answers being extracted from the document collection.

System and method for finding the most likely answer to a natural language question

Automated question answering is disclosed that relates to the selection of an answer to a question from a pool of potential answers which are manually or automatically extracted from a large collection of textual documents. The a feature extraction component, a feature combination component, an answer selection component, and an answer presentation component, among others, are included. The input to the system is a set of one or more natural language questions and a collection of textual document The output is a (possibly ranked) set of factual answers to the questions, these answers being extracted from the document collection.

This patent describes a novel system, method, and program product that are used in interactive natural language dialog. One or more presentation managers operating on a computer system present information from the computer system to one or more users over network interface(s) and accept queries from the users using one or more known input/output modalities (e.g. Speech, typed in text, pointing devices, etc.). A natural language parser parses one or more natural language phrases received over one or more of the network interfaces by one or more of the presentation managers into one or more logical forms (parsed user input), each logical form having a grammatical and structural organization. A dialog manager module maintains and directs interactive sessions between each of the users and the computer system. The dialog manager receives logical forms from one or more of the presentation managers and sends these to a taxonomical mapping process which matches the items of interest to the user against the content organization in the content database to match business categories and sends modified logical forms back to the dialog manager.

System, method and program product for interactive natural dialog

The present invention is a system, method, and program product that comprises a computer with a collection of documents to be searched. The documents contain free form (natural language) text. We define a set of labels called QA-Tokens, which function as abstractions of phrases or question-types. We define a pattern file, which consists of a number of pattern records, each of which has a question template, an associated question word pattern, and an associated set of QA-Tokens. We describe a query-analysis process which receives a query as input and matches it to one or more of the question templates, where a priority algorithm determines which match is used if there is more than one. The query-analysis process then replaces the associated question word pattern in the matching query with the associated set of QA-Tokens, and possibly some other words. This results in a processed query having some combination of original query tokens, new tokens from the pattern file, and QA-Tokens, possibly with weights. We describe a pattern-matching process that identifies patterns of text in the document collection and augments the location with corresponding QA-Tokens. We define a text index data structure which is an inverted list of the locations of all of the words in the document collection, together with the locations of all of the augmented QA-Tokens. A search process then matches the processed query against a window of a user-selected number of sentences that is slid across the document texts. A hit-list of top-scoring windows is returned to the user.

System, method and program product for answering questions using a search engine

Growing community of inventors

Ann Arbor, MI, United States of America

Dragomir Radkov Radev