CSAIL Publications and Digital Archive header
bullet Research Abstracts Home bullet CSAIL Digital Archive bullet Research Activities bullet CSAIL Home bullet

link to publications.csail.mit.edu link to www.csail.mit.edu horizontal line


Research Abstracts - 2007
horizontal line

horizontal line

vertical line
vertical line

Distributed Interpretation of Natural Language Requests

Gary Borchardt, Boris Katz, Federico Mora & Sue Felshin


When natural language requests—questions and commands—are to be processed in dynamic, distributed environments such as those involving mobile devices, a particular problem arises. It is often the case that natural language requests can only be understood—their ambiguities resolved—in the presence of specific, matching components of knowledge, and in distributed environments, this knowledge is distributed, requiring the networked devices and systems to collaborate not only toward the ultimate satisfaction of the received requests, but as well toward the initial understanding of those requests, so that it may be possible to satisfy them.


As part of the StartMobile effort, we have developed a representation language, called Moebius, that supports distributed interpretation and distributed fulfillment of natural language requests. Moebius serves to encode natural language requests at varied stages of syntactic and semantic interpretation, so that these requests may be relayed between systems—for instance, a user's mobile device, central servers, and other users' mobile devices—in order to receive additional interpretation and fulfillment. While Moebius specifically addresses the representation and processing of ambiguous requests, it is also applicable for more straightforward requests and thus we use the language as an interlingua for all requests received by StartMobile. Following is an example Moebius expression—depicting a substantially interpreted version of the English request "Remind my mother to take her medicine at 7pm.":

  • alert(object:person mother(of:person "user"), with:message_string "Take your medicine at 7pm.", at:time "2007-02-28T19:00:00")!

Moebius specifies syntactic relations loosely along the lines of a dependency tree representation (see, for example, [1]), and it adds semantic labels, drawn from a hierarchy of general to specific categories, to various elements within the representation. In using language itself as a representation, Moebius shares a common orientation with other representations such as ACE [2] and CPL [3]. However, whereas these approaches use simple language as a medium for translating knowledge into more formal representations, Moebius uses simple language for its own intrinsic benefits of being able to express and preserve the same sorts of ambiguities that arise in full-English requests. In general, ambiguity can arise from many sources—abstract verbs; syntactic ambiguities; ambiguous prepositions and conjunctions; abstract semantic categories; descriptive specifications of objects; ambiguous names, times and places; anaphora; and abstract adjectives and adverbs, to name a few—and the goal of Moebius is to capture a range of such ambiguities, in various stages of interpretation, in an English-like, structured format.

As a simple example of the use of Moebius to characterize an ambiguous request at different stages of interpretation, consider the request "Is John at IBM?". This question could be offered to ascertain whether or not John is employed by IBM, or it could be offered to determine whether or not John is, at the moment, physically present at an IBM facility. Which interpretation is intended may be construed by consulting the repertoire of capabilities offered by the device expected to fulfill the request—whether that device is known to be able to respond to one interpretation or the other—or, the interpretation may be construed by reference to contextual information from the current state of processing, or by consultation with a human user.

If the device that initially processes the request "Is John at IBM?" does not have access to the knowledge needed to fully interpret the request, then, using Moebius, that device can encode the request in a partially interpreted form:

  • is(subject:person "John", at:object "IBM")?

This representation parses the request syntactically, yet makes no commitment as to the semantic interpretation of the relationship between "John" and "IBM" or as to the specific semantic category of "IBM". If this request is relayed to another device or system that possesses the necessary knowledge to disambiguate the request, that system may cast the request into one of two more fully interpreted forms. If the determination is made that the request concerns physical presence at an IBM facility, the request can be re-expressed as

  • is(subject:person "John", at:facility "IBM")?

where "IBM" is classified semantically as a physical "facility". On the other hand, if the determination is made that the request concerns employment, the request can be re-expressed as

  • employs(subject:organization "IBM", object:person "John")?

where "IBM" is classified semantically as an abstract "organization" and the relationship is re-expressed as one of employment. Subsequent processing can proceed, then, according to the chosen interpretation.


Within the StartMobile effort, we are using Moebius to relay requests from the START information access system, which performs initial interpretation of received requests, to mobile devices charged with carrying out the requests. StartMobile currently handles a range of requests regarding the placing of phone calls, entering of reminders, retrieval of contact and calendar information, retrieval of text messages, taking pictures, retrieval of video tutorials, and other tasks. We are continuing to expand the coverage and use of the Moebius language as we add new request-handling capabilities to the StartMobile system.


This work is supported in part by the Nokia Corporation as part of the Nokia/MIT Lablet initiative.


[1] John Carroll, Ted Briscoe and Antonio Sanfilippo. Parser Evaluation: a Survey and a New Proposal. In Proceedings of the International Conference on Language Resources and Evaluation, 1998.

[2] Norbert E. Fuchs, Kaarel Kaljurand and Gerold Schneider. Attempto Controlled English Meets the Challenges of Knowledge Representation, Reasoning, Interoperability and User Interfaces. In Proceedings of the 19th International FLAIRS Conference (FLAIRS 2006), 2006.

[3] Peter Clark, Phil Harrison, Tom Jenkins, John Thompson and Rick Wojcik. Acquiring and Using World Knowledge using a Restricted Subset of English. In Proceedings of the 18th International FLAIRS Conference (FLAIRS 2005), 2005.


vertical line
vertical line
horizontal line

MIT logo Computer Science and Artificial Intelligence Laboratory (CSAIL)
The Stata Center, Building 32 - 32 Vassar Street - Cambridge, MA 02139 - USA
tel:+1-617-253-0073 - publications@csail.mit.edu