Teletandem is a model of virtual exchange in which two speakers of different languages meet over videoconferencing tools so that they can learn each other's language and culture. Research and practice on teletandem draw on the principles of autonomy (each participant should make decisions about their own learning), reciprocity (each participant should assist their partner's learning), and separation of languages (each language should have its own moment of practice and should not be mixed). Some researchers, however, do not consider separation of languages to be a principle of tandem learning and few studies have challenged it from the theoretical perspective of code-switching (Picoli & Salomão, 2020) and investigated learner interaction using translanguaging (Canals, 2021; Satar, 2020; Walker, 2017). In this paper, data come from the MuLTeC teletandem corpus. Using social semiotics, we investigate six recorded videoconferencing sessions of a Brazilian/American dyad with low language proficiency level and explore learners' transformative engagement with signs (Bezemer & Kress, 2016). Results reveal that participants draw on all the resources in their complete language repertoire (García & Wei, 2014) through translingual and transmodal practices, which allows them to make meanings beyond what they can express within their target language proficiency, creating opportunities to extend their semiotic repertoires. The findings evidence the inappropriateness of the separation of languages principle which neither works in practice nor is useful within current perspectives in language learning and teaching. Therefore, we propose a reframing of the principle as the translanguaging principle: based on the notion of translanguaging as co-learning, each participant should play fluid roles of expert/learner and capitalise on opportunities to extend each other's semiotic repertoires within multilingual and multimodal interaction.