Proposal Status Inquiry: Panel Summary #1 Details of Proposal 0205223 Panel Summary #1 PROPOSAL NO.: 0205223 PANEL SUMMARY: The proposal does not explain precisely what this proposed program of research adds to the results the collaborators have already obtained in their previous and on-going work, or the interaction/synergy/overlap with the project(s) supported by European funds, such as Nespole; nor does it clearly articulate the anticipated impact of the research. PANEL RECOMMENDATION: Not Competitive Please use the back navigation button to return to the previous page Proposal Status Inquiry: Review #1 Details of Proposal 0205223 Review #1 PROPOSAL NO.: 0205223 INSTITUTION: Carnegie Mellon University NSF PROGRAM: ITR MEDIUM (GROUP) GRANTS PRINCIPAL INVESTIGATOR: Lavie, Alon TITLE: ITR: SPECTRUM - Speech Communication and Translation Under Mobile Environments RATING:Good REVIEW: What is the intellectual merit of the proposed activity? 0.Key effort: wireless help desk application. 1. It is cheering (though a bit ironic) to see W. Woods' 1970s BBN HWIM SLS system ("Hear What I Mean") being revived. Woods *did* mean what he said: the system in fact performed better than existing systems of that time in utterance-to-meaning/speaker intention mappings - though it did worse on individual word recognition. Better late than never. There are concerns. Consider semantic grammar induction. p. 9: "we plan first to investigate the extent to which grammar rules can be learned completely automatically" Puzzling: one might have thought the lessons from the last few decades of work here is that this is very very difficult indeed. In general, it cannot be done 'completely automatically', so why investigate this first? Unless somebody has cracked the child language acquisition problem in an unforeseen way. 3. Dialog monitor and modality management. The proposal notes (correctly, in my view) that identifying speech acts did not help as much as first thought Appendix proposes to recover emotional state through prosody. This has been a constant lure in the field. Do we have any hard evidence that it is really possible? Motherese experiments show that adults don't attend to disambiguation of their own sentences unless they are actually *aware* of that ambiguity - which they are not, sometimes. See Snedeker, J. (2000b). Combining cues: An analysis of the input to verb learning. Paper presented at the Stanford Words Conference, Stanford, CA, April 2000. 4. Comment on management. This is a *very* complex operation across countries. Special care should be taken here, obviously. What are the broader impacts of the proposed activity? Summary Statement 1. This proposal should be funded, if only to provide a 'control case' against which to test the past quarter century of investment (now evidently at asymptote) in non-HWIM based systems. Please use the back navigation button to return to the previous page Proposal Status Inquiry: Review #2 Details of Proposal 0205223 Review #2 PROPOSAL NO.: 0205223 INSTITUTION: Carnegie Mellon University NSF PROGRAM: ITR MEDIUM (GROUP) GRANTS PRINCIPAL INVESTIGATOR: Lavie, Alon TITLE: ITR: SPECTRUM - Speech Communication and Translation Under Mobile Environments RATING:Good REVIEW: What is the intellectual merit of the proposed activity? This contribution claimed for this project is incremental: the project improves upon the collaborators’ previous project. The goals are clearly and precisely defined, as are the proposed strategies for their realization. The project does not promise a ground-shifting advance but an increase in the robustness and accuracy. While a number of strategies are proposed to solve technical problems in speech recognition, the most significant and innovative aspects of the project is in the project are in the area of usability. The international consortium assembled for this project is an enormous asset. What are the broader impacts of the proposed activity? The potential cultural impact of a robust and usable mobile system for translation is tremendous. (The potential impact is described more effectively in the attached European proposal than in the narrative for NSF.) This impact will be severely limited, however, if the technologies developed are too domain-specific. The research focuses on a very specific, limited test case: a help-desk for tourists. The proposed system exploits the limited range of likely communicative goals within this context. The proposal does not clearly indicate how the solutions developed here can be generalized to more open-ended situations. Summary Statement Please use the back navigation button to return to the previous page Proposal Status Inquiry: Review #3 Details of Proposal 0205223 Review #3 PROPOSAL NO.: 0205223 INSTITUTION: Carnegie Mellon University NSF PROGRAM: ITR MEDIUM (GROUP) GRANTS PRINCIPAL INVESTIGATOR: Lavie, Alon TITLE: ITR: SPECTRUM - Speech Communication and Translation Under Mobile Environments RATING:Good REVIEW: What is the intellectual merit of the proposed activity? The present proposal seeks to extend the work of the authors to further develop the robustness of communication in spoken language translation (SLT) system for E-commerce - currently supported by a $1.5 million 3-year grant, referred to as the NESPOLE! Project, that was awarded by NSF. The authors are also seeking funding through a parallel proposal for the Spectrum project that has been submitted to the EC (IST-2001-34738). Spectrum project is a cooperative venture of Carnegie Mellon and the three European research groups at ITC-irst in Italy, UJF in France and the University of Karlsruhe in Germany. This proposal submitted to the NSF-ITR Program requests funding for Carnegie Mellon's participation in the project. However, it is not clear what is part of the NESPOLE! project and what is part of the present proposal. Will the work described in this proposal not get done if NSF-ITR does not fund it? What are the broader impacts of the proposed activity? The Spectrum project is an ambitious project that will contribute to the science of speech recognition and machine translation even though the proposal often reads it was describing the need for additional programming efforts rather than contributing to testing theories. Summary Statement The proposed system seems to be a well thought out, with an orderly progression of work for advancing spoken language translation systems. The proposal seems is broad in its scope proposing to develop and test multiple strategies for 1) decoding acoustic signals from a variety of sources (emphasis on mobile phones); 2) developing different approaches that are based on machine-learning to significantly reduce the amount of effort required in expanding the coverage of semantic grammars to broader or new domains; and 3) developing a dialogue monitoring component to reliably detect failures in the communication and their causes thus assisting the user in overcoming communication difficulties by using alternative modalities. Please use the back navigation button to return to the previous page Proposal Status Inquiry: Review #4 Details of Proposal 0205223 Review #4 PROPOSAL NO.: 0205223 INSTITUTION: Carnegie Mellon University NSF PROGRAM: ITR MEDIUM (GROUP) GRANTS PRINCIPAL INVESTIGATOR: Lavie, Alon TITLE: ITR: SPECTRUM - Speech Communication and Translation Under Mobile Environments RATING:Good REVIEW: What is the intellectual merit of the proposed activity? Engineering effort in spoken language translation. Interesting application: a prototype of a mobile platform for travel-related services including spoken human-machine dialogue. What are the broader impacts of the proposed activity? Robustness in communication man-machine has a number of important applications. It is an engineering challenge more suited to collaboration between Universities and companies (as in the European community)than just an academic environment. Summary Statement The proposal is interesting. None of the PIs is a faculty member (?) but some of them have good background and good credentials in the area (for instance Alex Waible). The institutions involved are good (I know IRST and I rate it high in Europe). The proposal has some unusual aspects: it seems that European funding is also sought for part of it or for a related version of the project; and its overlap with another project (NESPOLE!) started in 2000 is unclear to me and should be discussed in the panel. Please use the back navigation button to return to the previous page