Integrated language model, related systems and methods
An integrated language model includes an upper-level language model component and a lower-level language model component, with the upper-level language model component including a non-terminal and the lower-level language model component being applied to the non-terminal. The upper-level and lower-level language model components can be of the same or different language model formats, including finite state grammar (FSG) and statistical language model (SLM) formats. Systems and methods for making integrated language models allow designation of language model formats for the upper-level and lower-level components and identification of non-terminals. Automatic non-terminal replacement and retention criteria can be used to facilitate the generation of one or both language model components, which can include the modification of existing language models.
1. A method of making an integrated language model for a speech recognition engine, the method comprising:
using at least one processor and machine readable memory:
identifying a first language model format for an upper-level language model component;
identifying a plurality of text elements to be represented by a non-terminal in the upper-level language model component;
generating the upper-level language model component including the non-terminal;
identifying a second language model format for a lower-level language model component to be applied to the non-terminal of the upper-level language model component; and
generating the lower-level language model component;
wherein generating the upper-level language model component includes modifying an existing language model; and
wherein identifying the plurality of text elements to be represented by the non-terminal in the upper-level language model component includes applying an automatic text element replacement criterion to the existing language model.
2. The method of claim 1 , wherein the first language model format is a finite state grammar format.
3. The method of claim 2 , wherein the second language model format is a statistical language model format.
4. The method of claim 1 , wherein a plurality of non-terminals are included in the upper-level language model.
5. The method of claim 4 , wherein a plurality of lower-level language model components are applied to the plurality of non-terminals.
6. The method of claim 5 , wherein the plurality of lower-level language model components include at least two language model components having different language model formats.
7. The method of claim 5 , further comprising identifying to which of the plurality of non-terminals each of the plurality of lower-level language models is to be applied during operation of the speech recognition engine.
8. The method of claim 1 , wherein the lower-level language model component also includes a non-terminal, and the method further comprises identifying a third language model format for an additional lower-level language model component to be applied to the non-terminal of the lower-level language model component, and generating the additional language model component.
9. The method of claim 1 , wherein the existing language model is a finite state grammar format language model.
10. The method of claim 1 , wherein the existing language model is a statistical language model format language model.
11. The method of claim 1 , wherein the text element replacement criterion is to replace text element sequences having a definable length.
12. The method of claim 11 , wherein the definable length is a number of digits.
13. The method of claim 1 , wherein the text element replacement criterion is to replace text element sequences having a definable value range.
14. The method of claim 1 , wherein a plurality of automatic text element replacement criteria are applied.
15. The method of claim 1 , wherein identifying the plurality of text elements to be represented by the non-terminal in the upper-level language model component further includes applying an automatic text element retention criterion to the existing language model.
16. The method of claim 1 , wherein generating the lower-level language model component includes modifying an existing language model.
17. The method of claim 16 , wherein modifying the existing language model includes automatically eliminating text elements that are determined not to be relevant to the non-terminal, the determination being based on a non-terminal replacement criterion applied to identify the plurality of text elements to be represented by the non-terminal.
18. The method of claim 1 , further comprising generating instructions for a speech recognition engine decoder to apply the upper-level and lower-level language model components during operation.
19. The method of claim 1 , wherein the text elements are words.