How Speech Act Analysis Can Unlock Robot Emotions

3.0 Exploration 2: Sentence Speech Act Analysis

What is a speech act In the previous part we noted that studying the communicative function of utterances is essentially studying dialogue speech acts. A speech act can be analyzed from three angles: the locutionary act (the act of speaking), the illocutionary act (the communicative function such as asking a question, giving a command, or expressing an attitude), and the perlocutionary act (the effect on the listener, e.g., closing a window after hearing “I’m cold”).

Current limitations: judgments are based only on the literal text, without incorporating speaker facial expressions, tone, or broader context. For example, “你真聪明” can be a sincere compliment or sarcastic, depending on tone and facial cues. Future multimodal inputs will enable more precise analysis.

Classification of Speech Acts

From a communicative‑function perspective, all utterances can be divided into six categories, based on the theories of Searle and Professor Wu Jianfeng: inquiry, command, declaration, promise, expression, and statement. Expression can be further subdivided into more specific behaviors.

High‑Frequency Sentence Speech‑Act Analysis

Using data from the Dingtang smart screen, we classified frequent user utterances in casual‑talk scenarios.

Findings

Inquiry, expression, and command are the three most frequent speech‑act types.

The classification covers the entire set of user utterances.

Each speech‑act type can be linked to a specific logic for inferring robot emotion.

Conclusion

Analysis of real high‑frequency user utterances shows that every sentence corresponds to a speech act, and sentences of the same act share syntactic patterns and regularities. The act determines the semantic focus for the robot’s response: in inquiries the topic is the focus, in commands the action to be performed is the focus, and in statements the described event is the focus. Consequently, the identified focus becomes the key factor influencing robot emotion.

While sentence patterns give clues (e.g., questions often indicate inquiry, imperatives indicate command, exclamations indicate expression), they do not map one‑to‑one with speech acts. Human judgment is still required to disambiguate cases such as “你怎么这么好看？” (a question that actually expresses praise) or “我喜欢你” (a declarative sentence that conveys a confession).

Relation Between Speech Acts and Propositional Content

After defining speech‑act categories, we examined how the same propositional content can be expressed with different acts. For example:

“你吃过麦当劳吗？”

“你要去吃麦当劳。”

“你马上过来吃麦当劳。”

All three discuss the action “you eat McDonald’s,” but the first is an inquiry, the second a prediction, and the third a command. The robot’s emotional response varies accordingly: answering the question may involve sharing personal experience, answering the prediction involves confirming intent, and answering the command involves expressing willingness or refusal.

Emotion Response Mechanism Generation Formula

The explorations make configurable robot emotion possible. Based on the analysis we devised detailed emotion‑inference rules and generated a configurable formula, allowing creators to customize robot reply style and emotional tone.

The first dropdown selects the speech act, the second selects the propositional content, and the third selects the robot’s expressive action or facial expression.

Semantic Annotation

Accurate semantic annotation is crucial for the emotion configuration. Different speech‑act contexts require different annotation dimensions, such as topic, event nature, or robot capability. For instance, statements describing “user harm” (e.g., “我脚崴了”, “我考砸了”) are grouped under a “user harm” category, enabling the robot to respond with appropriate emotions like comfort, concern, or sarcasm.

Conclusion

The goal of emotion‑system research is to give robots logical, personality‑driven artificial emotions that enhance user experience. While current methods provide shallow emotional interaction, future advances may enable deeper, more human‑like emotional conversations.