This paper explores face-to-face interaction among Taiwanese, Indonesian, and Indian speakers employing a multimodal corpus linguistics approach to investigate semantic categories of speech that most often co-occur with gestures, and whether the gesture-speech link is to a certain degree affected by a speaker's English proficiency levels or language/culture backgrounds. It is shown that speech most commonly co-occurs with gestures in the categories of moving, coming and going, general objects, numbers, location and direction, and time. The findings show similar predilections of gesture-speech production by speakers despite distinct cultural and linguistic backgrounds. The gesture-speech link was shown to fall into six discrete categories: reinforcing, integrating, supplementary, complementary, contradictory, and others. While findings display that the gesture-speech link is not significantly affected by distinct language backgrounds of a speaker, speakers of a high proficiency level tended to use significantly more gestures that serve reinforcing and integrating functions, whereas less proficient speakers produced more gestures as complements and other gestures that have no obvious relationship to the conceptual content of their accompanying speech.