
NSF Org: |
IIS Division of Information & Intelligent Systems |
Recipient: |
|
Initial Amendment Date: | May 6, 2015 |
Latest Amendment Date: | May 6, 2015 |
Award Number: | 1459300 |
Award Instrument: | Standard Grant |
Program Manager: |
D. Langendoen
IIS Division of Information & Intelligent Systems CSE Directorate for Computer and Information Science and Engineering |
Start Date: | May 1, 2015 |
End Date: | April 30, 2019 (Estimated) |
Total Intended Award Amount: | $174,485.00 |
Total Awarded Amount to Date: | $174,485.00 |
Funds Obligated to Date: |
|
History of Investigator: |
|
Recipient Sponsored Research Office: |
201 OLD MAIN UNIVERSITY PARK PA US 16802-1503 (814)865-1372 |
Sponsor Congressional District: |
|
Primary Place of Performance: |
316D IST Building University Park PA US 16802-6823 |
Primary Place of
Performance Congressional District: |
|
Unique Entity Identifier (UEI): |
|
Parent UEI: |
|
NSF Program(s): | Robust Intelligence |
Primary Program Source: |
|
Program Reference Code(s): |
|
Program Element Code(s): |
|
Award Agency Code: | 4900 |
Fund Agency Code: | 4900 |
Assistance Listing Number(s): | 47.070 |
ABSTRACT
Language use in real-world dialogue happens in context. Linguistic choices depend on previous ones: for example, the chosen words and sentence structures tend to mirror what was used previously by a conversation partner. This subtle adaptation process has been called "alignment". Alignment appears to help people understand each other in dialogue, and it seems to extend to human-computer interfaces, too. The concrete functions of alignment in dialogue are, however, unclear. Is it merely a useful epiphenomenon of how human memory works? Does it serve as a social or communicative signal? Is it indicative of a person's empathy? Does it help communities find a common language over long periods of time? Recent work has established that one of the consequences of alignment is persistent language change in the individual. There also is preliminary evidence that over time, groups of people talking to one another will converge in their choice of words and sentence structure. In other words, they find a common language. The project will devise computational models that describe and quantify these processes. With these, one can detect them in actual language use, such as in web-forums. In fact, the project will use big datasets from decades of web-forum messages to produce those models. The computational models will explain and predict processes in a way that makes them exploitable in modern social networks as well as for data science. Consider the example of a web-forum that connects those suffering from a disease so they can lend each other emotional and informational support. The models can detect and predict which messages in this web-forum are most supportive on the intended level, and whether they align to the person asking a question. A possible application of this may improve web-forum discourse by prioritizing search results and by making reading suggestions. Alignment models may also improve analysis techniques for large datasets by spotting networks of mutual supporters.
Models will be created in order to describe and explain alignment and language change in natural-language dialogue. The models will be computational and statistical to allow for exploitation of interactive alignment in natural-language dialogue as a feature in social network applications. Statistical alignment models describe language change over time as a function of variables that characterize the individual's behavior, memory, and of network information. These models will be fitted to longitudinal datasets derived from web-based, topic-oriented conversation threads. At the individual level, they will help refine cognitive-computational models of memory function in language production, which will be constrained by the well-validated ACT-R framework. The viability of the approach is supported by preliminary work on corpus-based syntactic priming and ACT-R models of language production, and pilot experiments showing alignment in the corpus. The outcomes of the project may point to novel methods of prioritizing and filtering the most helpful content and can address quality of life and well-being of patients such as those of the peer-support community whose conversations were studied in the investigator's work motivating the proposal.
PUBLICATIONS PRODUCED AS A RESULT OF THIS RESEARCH
Note:
When clicking on a Digital Object Identifier (DOI) number, you will be taken to an external
site maintained by the publisher. Some full text articles may not yet be available without a
charge during the embargo (administrative interval).
Some links on this page may take you to non-federal websites. Their policies may differ from
this site.
PROJECT OUTCOMES REPORT
Disclaimer
This Project Outcomes Report for the General Public is displayed verbatim as submitted by the Principal Investigator (PI) for this award. Any opinions, findings, and conclusions or recommendations expressed in this Report are those of the PI and do not necessarily reflect the views of the National Science Foundation; NSF has not approved or endorsed its content.
The project helped us understand how people strategically distribute information across their conversations. We studied how these strategies can make conversation more or less effective, leading to successful outcomes in collaborative tasks. Using existing datasets in English and Danish (of task-oriented dialogue), we identified features that analyze the periodic changes in information density between two interlocutors. Information density is defined as how well a computer model can predict words in each sentence, given the words that were used up to that point: the level of "surprisal" experienced when hearing a word is indicative of information. We find that in successful communication, interlocutors take turns in contributing information, and that increasing information by an interlocutor responding to a topic can indicate effective understanding of what is conveyed. This project developed computer models that can automatically analyze this and, thus, automatically determine how effective a given conversation is. The project also examined adaptation effects in conversations. In addition to the prepared English and Danish datasets, we used dialogue datasets acquired from internet forums, which proved to be a rich source of language data.
The work under this grant identified important features of effective communication, which is now being applied to develop industrial human-computer dialogue systems.
Last Modified: 09/30/2019
Modified by: David T Reitter
Please report errors in award information by writing to: awardsearch@nsf.gov.