Helios
From Olympus
The Helios module integrates multiple sources of information to determine the confidence of a particular understanding generated by the Phoenix semantic parser, and presents a unified assessment of input understanding to the dialog management system. Helios was originally conceived and implemented by Paul Constantinides, as part of the CMU Communicator project.
The following features are computed from a history of parses:
| feature name | description |
|---|---|
| parse_str | the parsed string |
| top_slots | the top-level slots |
| slot_num | the number of slots in the parse |
| slot_num_gt_1 | |
| slot_num_gt_2 | |
| new_slots_num | the number of new slots in the parse |
| new_slots_num_bool | the presence of new slots in the parse |
| rep_slots_num | the number of repeated slots in the parse |
| rep_slots_num_bool | the presence of repeated slots in the parse |
| total_num_parses | the total number of parses (all hyps) |
| hyp_num_parses | the number of parses for this hypothesis |
| num_parses_ratio | the ratio of number of parses |
| uncov_num | the number of words not covered by the parse |
| uncov_num_bool | |
| uncov_num_gt_1 | |
| uncov_ratio | the ratio of uncovered words |
| uncov_ratio_gtm | |
| frag_num | the number of fragments in the parse |
| frag_ratio | the ratio of fragments |
| frag_ratio_gtm | |
| gap_num | the number of gaps in the parse |
| gap_num_bool | |
| frag_and_gap_num | the total number of fragments and gaps |
| frag_and_gap_num_gt_1 | |
| frag_mode | the fragmentation mode |
| h_avg_gap_num | the average number of gaps so far in the dialog |
| h_avg_uncov_num | the average number of uncovered so far in the dialog |
Publications
- Bohus, D. & Rudnicky, A. 2002, November. Integrating multiple knowlege sources for utterance-level confidence annotation in the CMU Communicator spoken dialog system. (Tech. Rep. No. CMU-CS-02-190). Pittsburgh, Pennsylvania: School of Computer Science, Carnegie Mellon University.
- Zhang, Rong & Alexander Rudnicky (2001), "Word Level Confidence Annotation using Combinations of Features", Eurospeech
