| P02-1025 | information and with the same | SLM training | procedure we can estimate the |
| W07-0311 | 4.3 Amount of training data for | SLM training | For the remaining Sections , |
| E06-1046 | free deletion of any words in the | SLM training | data which are not found in the |
| P02-1025 | Section 2.2 and Section 3 . Because | SLM training | on the 20M words of WSJ text |
| P02-1025 | the N-best EM variant used for | SLM training | algorithm increases the likelihood |
| W07-0303 | combined grammar by composing an | SLM training | corpus based on the slide content |
| W07-0303 | command phrases is balanced in the | SLM training | data . The resulting combined |
