Language model prediction of API call calls and verbal responses

A method (500) includes obtaining an utterance (106) from a user (10) for a digital assistant (50). The method includes generating a first predicted string (214) based on the utterance using a language model (210) and determining whether the first predicted string includes an application programming interface (API) call (222) to retrieve a program (60) via an API (62). When the first predicted string includes an API call, the method includes invoking a program via the API to retrieve a program result (224); receiving a program result; updating the dialog context comprising the utterance using the program results (250); and generating a second predicted string (214) based on the updated dialog context using the language model. When the first predicted string does not include the API call, the method includes providing an utterance response to the utterance based on the first predicted string (226). 一种方法(500)包括获得来自用户(10)的针对数字助理(50)的话语(106)。所述方法包括使用语言模型(210)基于话语生成第一预测字符串(214)，并确定第一预测字符串是否包括经由应用编程接口(API)(62)调取程序(