Writing a general grammar

[SmillingMan]

Hi,

I want to write a grammar that accpets all the words in a given dictionary. Can you please help me with that. I think it should be a one line grammer, something like

public <newGrammar> = *;

Thanks,
Foad

[Nickolay V. Shmyrev]

It can be unigram language model I suppose

[SmillingMan]

Yes, it's a unigram (a list of single words) but how can I write the grammar file so that all the words are accepted?

Thanks,
Foad

[Nickolay V. Shmyrev]

Something like:

\data\
ngram 1=8

\1-grams:
-1.1244 <UNK> 0.0000
-1.4254 5.75 -0.4605
-1.7264 </s> 0.0000
-1.4254 <s> -0.4605
-1.4254 Fed -0.4605
-1.4254 Le -0.4605
-1.4254 a -0.4605
-1.4254 abaissé -0.4605

it's an arpa format. You can use equal weights or cmulmtk tools to create arpabo model from text with proper probabilities.

[SmillingMan]

Thanks for the sample. Is there a way to write a JSGF grammar to accept the general list?

To recognize two words (one, two), I can write a JSGF rule:

public <newGrammar> = one | two;

I am looking for a similar rule to accept all words in the dictionary.

Best,
Foad

[Nickolay V. Shmyrev]

Hm, * denotes repeating. So the thing you like should be addition to JSGF standard I suppose. I think it even can be implemented with a special tag like <WORD>. Tag like <VOID> is similar but I'm not sure how it works. When you will have too much words in a dictionary, such thing will not be useful in my opinion since it will not constrain search at all.

[David Huggins-Daines]

This is an interesting idea, I think that it should be implemented in the FSG search module. It would not be useful to have a grammar that simply consists of <WORD>, but it would be nice to be able to allow any word in the dictionary to be recognized at certain points (perhaps as a "backoff" node with low probability).