Interpreting expressions having potentially ambiguous meanings in different domains
The present invention extends to methods, systems, and computer program products for interpreting expressions having potentially ambiguous meanings in different domains. Multi-domain natural language understanding systems can support a variety of different types of clients. Expressions can be interpreted across multiple domains. Weights can be assigned to domains. Weights can be client specific or expression specific so that a chosen interpretation is more likely correct for the type of client or for its context. Stored weight sets can be chosen according to identifying information carried as metadata with expressions or weight sets carried directly as metadata. Domains can additionally or alternatively be ranked in ordered lists or comparative domain pairs of to favor some domains over others as appropriate for client type or client context.
1. A method comprising: providing a configurable platform for natural language understanding for various types of clients:
receiving, at the configurable platform and from developer input into a user interface, instructions to apply, for a client type identifier, a first numerical weight value to a first domain and a second numerical weight value to a second domain;
receiving an expression package containing a natural language expression and the client type identifier;
interpreting the natural language expression according to the first domain to produce a first interpretation and a first score for the first domain;
interpreting the natural language expression according to the second domain to produce a second interpretation and a first score for the second domain;
using the client type identifier to identify the first weight and to identify the second weight;
applying the first numerical weight value to the first score for the first domain to compute a second score for the first domain;
applying the second numerical weight value to the first score for the second domain to compute a second score for the second domain;
selecting the first interpretation as the appropriate interpretation of the natural language expression in view of the second score for the first domain and the second score for the second domain; and
computing a response to the natural language expression based on selection of the first interpretation.
2. The method as recited in claim 1 , wherein selecting the first interpretation as the appropriate interpretation comprises selecting the first interpretation based on the second score for the first domain being higher than the second score for the second domain.
3. The method as recited in claim 1 , wherein selecting the first interpretation as the appropriate interpretation comprises selecting the first interpretation based on the second score for the first domain exceeding a threshold.
4. The method as recited in claim 1 , wherein receiving an expression package comprises receiving the expression package from a client device, and wherein the expression package further indicates which application is running in the foreground at the client device.
5. The method as recited in claim 4 , further comprising taking a computerized action with respect to the client device to implement the computed response.
6. A method comprising:
receiving a client defined expression package from a client device, the client defined expression package containing a natural language expression, a first numerical weight value, and a second numerical weight value, the first numerical weight value and the second numerical weight value defined to apply to interpretations of the natural language expression in a second domain;
interpreting the natural language expression according to the first domain to produce a first interpretation and a first score for the first domain;
interpreting the natural language expression according to the second domain to produce a second interpretation and a first score for the second domain;
applying the first numerical weight value to the first score for the first domain to compute a second score for the first domain;
applying the second numerical weight value to the first score for the second domain to compute a second score for the second domain;
selecting the first interpretation as the appropriate interpretation of the natural language expression in view of the second score for the first domain and the second score for the second domain; and
computing a response to the natural language expression based on selection of the first interpretation.
7. The method of claim 6 , wherein receiving an expression package comprises receiving an expression package wherein the first numerical weight value is tagged with an identifier for the first domain and the second numerical weight value is tagged with an identifier for the second domain.
8. The method of claim 6 , wherein receiving an expression package comprises receiving an expression package containing a natural language expression and an array of domain weights.
9. The method as recited in claim 6 , wherein selecting the first interpretation as the appropriate interpretation comprises selecting the first interpretation based on the second score for the first domain being higher than the second score for the second domain.
10. The method as recited in claim 6 , wherein selecting the first interpretation as the appropriate interpretation comprises selecting the first interpretation based on the second score for the first domain exceeding a threshold.
11. The method as recited in claim 6 , wherein receiving an expression package from a client device comprises receiving an expression package from a client device selected from among: a smart speaker, a mobile phone, and an automobile.
12. The method as recited in claim 11 , further comprising taking a computerized action with respect to the client device to implement the computed response.
13. A method comprising:
receiving an expression package containing a natural language expression and indicating a domain rank list defining a
preference for natural language expression interpretations from a first domain relative to natural language expression interpretations from a second domain;
interpreting the natural language expression according to the first domain to produce a first interpretation and a first score for the first domain;
interpreting the natural language expression according to the second domain to produce a second interpretation and a second score for the second domain;
determining that the first score exceeds a first threshold, the second score exceeds a second threshold, and the second score is higher than the first score;
selecting the first interpretation as the appropriate interpretation of the natural language expression in view of the defined preference in the domain rank list and based on the first score exceeding the first threshold and notwithstanding that the second score is higher than the first score; and
computing a response to the natural language expression based on selection of the first interpretation.
14. The method as recited in claim 13 , wherein accessing a domain rank list comprises receiving the domain rank list along with the natural language expression.
15. The method as recited in claim 13 , wherein receiving an expression package comprises receiving an expression package from a client device, the client device selected from among: a smart speaker, a mobile phone, and an automobile.
16. The method as recited in claim 15 , further comprising taking a computerized action with respect to the client device to implement the computed response.
17. The method of claim 13 , further comprising:
providing a configurable platform for natural language understanding for various types of clients; and
receiving, at the configurable platform and from developer input into a user interface, instructions to give the preference to interpretations from the first domain relative to interpretations from the second domain.
18. The method of claim 13 , wherein receiving an expression package comprises receiving a client defined expression package that includes the domain rank list from a client device, the domain rank list defining domain preferences among a plurality of domains, including the first domain and the second domain, relative to one another and to be applied to interpretations of the natural language expression.
19. The meth of claim 13 , wherein selecting the first interpretation as the appropriate interpretation comprises overriding selection of second interpretation even though the second score is higher than the first score.
20. The meth of claim 13 , wherein selecting the first interpretation as the appropriate interpretation comprises determining that the second domain is not included in the domain rank list.