1. 1 Boostrapping language models for dialogue systems Karl Weilhammer, Matthew N Stuttle, Steve Young Presenter: Hsuan-Sheng Chiu

2. 2 Introduction Poor speech recognition performance is often described as the single most important factor prohibiting the wider use of spoken dialogue systems Many dialogue systems use recognition grammars instead of statistical language models In this paper, they compare the standard approach with using SLMs that trained from different corpora

3. 3 Experimental Setup Collection of example utterences –Ask 9 co-researchers who were familiar with the task to submit a set of 10 “example” interactions with the system and a number of more advanced dialogues –The training set was mostly used as held out data for interpolation, selection of the best model and other purposes –The test set was used for all the test runs done in the tourist information domain

4. 4 Experimental Setup (cont.) Generation and recognition grammar –A simple HTK grammar was written consisting of around 80 rules in EBNF and used in two ways it was converted into a word network (16996 nodes and 40776 transitions) A corpus of random sentences was generated from it –The task grammar was structured 1. task specific semantic concepts (prices, hotel names, …) 2. General concepts (local relations, numbers, dates, …) 3. Query predicates (Want, Find, Exists, Select, …) 4. basic phrases (yes, No, DontMind, Grumble, …) 5. List of sub-grammars for user answers to all prompts 6. Main Grammar

5. 5 Acoustic Models Use trigram decoder in Application Toolkit for HTK Acoustic models: –Training: WSJCAM0, internal triphone models (92 speakers of British English, 7900 read sentences, 130k words) –Adpatation: SACTI, 43 users, 3000 sentences, 20k words) –MAP and HLDA

6. 6 Experiments Baseline: use generation grammar which was compiled into a recognition network and yielded an WER of 40.4% Grammar networks vs. statistical language models –Use HTK-random sentence generator to generate a corpus –Relative improvement of 29% 30 A B A. Random sentence generator overweights common n-gram by repeating short sentencs more often than long sentences B. WERs are less sensitive to the size of the corpus

7. 7 Experiments (cont.) In-domain language model –WOZ Corpus –A WOZ corpus in the Tourist information domain (SACTI) was collected –The major part of the dialogues was recorded in a simulated automated speech recognition channel –Half of the corpus consists of speech only dialogues –In the other half an interactive map interface could also be used along with speech

8. 8 Experiments (cont.) Ideal dialogue data for statistical methods –Human-human (HH) channel Symmetric, prosodic information, few “recognition” error –ASR channel (Human-Computer (HC) ) Using a end-pointer, eliminate the prosodic information –It is not desirable for the collection framework to use a fixed or random policy A policy is a mapping from states to system actions Modified framework Simulated ASR channel

9. 9 Experiments (cont.) Wizard of Oz Experiments

10. 10 Experiments (cont.) The problem with this corpus is that it was recorded well before the dialogue system was defined that was assumed for the collection of the test set Given the near-identical WER, it is much faster and cheater to adapt a simple grammar than to record and transcribe in-domain data

11. 11 Experiments (cont.) Combining grammar and “in-domain” language models The perplexity on the held out data is a good predictor for the test set WER The interpolation of grammar corpus with a corpus with a natural speech straightens out most of the artifacts caused by the random generator

12. 12 Experiments (cont.) Interpolation language models derived from a grammar and a standard corpus –Fisher Corpus contains transcriptions of conversations about different topics –The idea behind this is that the grammar generated data will contribute in-domain n-grams and the general corpus will add colloquial phrases –For the reasons of comparison, the vocabulary used to build the Fisher SLM contained all words of the grammar and WOZ data collection – The absolute WER improvement that would have been selected based on perplexity minimum at the held out set is only 0.25%

13. 13 Experiments (cont.) Sentence selection using perplexity filtering –Collect a corpus of 2.4 M words (188,909 sentences) from Internet –It might be rewarding to extract all relevant sentences of the corpus and give them a higher weight than the remaining sentences –Build a LM from a seed corpus LM Seed –Build a LM from the large corpus LM Large –Calculate for each sentence and sort corpus according to PP Rel –Select n lines with lowest PPRel and build a LM LMSelected –Do the same with the remaining lines –Interpolate LM Seed, LM Selected and LM NotSelected

14. 14 Experiments (cont.) –LM Seed : 10M grammar generated sentences –Second round of LM Seed : interpolated Grammar-Fisher SLM Interpolating with the web corpus results in relative improvement of 3% Applying ppl filtering improved WER by another 2% relative

15. 15 Experiments (cont.) After a noisy initial segment (one-word sentences), the perplexity stabilizes at a certain level

16. 16 Conclusions Compare speech recognition results using a recognition grammar with different kinds of SLMs 1. Corpus trained on 30k sentences generated by grammar with WER by 29% relative 2. In-domain corpus with 21% relative improvement 3. Standard Corpus with Sentence Filtering Effective language models can be built for boot-strapping a spoken dialogue systems without recourse to expensive WOZ data collections