Effect of competence perception on lexical alignment in HCI

People repeat each other words in conversation, for example by using the word "brolly" instead of the word "umbrella" after their interlocutor used "umbrella". In other words, people entrain at a lexical level (Branigan et al., 2011). Lexical entrainment has been described as a consequence of alignment at the level of lexical representations, based on priming (Pickering and Garrod, 2004) but also as an audience design mechanism (e.g. speakers may repeat their interlocutors' words to facilitate their comprehension, in particular when their linguistic competence is assumed not to be optimal) (Brennan and Clark, 1996) or a combination of the two (Branigan et al.,2011). Speakers entrain with their interlocutors regardless of who they are. However, some elements of interlocutors' identity can modulate entrainment: Not only do people repeat their interlocutors' words when they interact with other people, but also when they interact with computer systems (such as virtual agents or social robots) (Branigan et al.,2011; Cirillo et al., 2021). Branigan et al. (2011) showed that the tendency to entrain is modulated by the perceived competence of the interlocutors (i.e. speakers entrain more often with systems that are perceived as less competent), which supports the audience design explanation or the combination between audience design and priming. However, it is still not clear how automatic mechanisms (based on priming) and strategic mechanisms (based on audience design) interact. Ivanova et al. (2020) proposed that attention might have a role: People might pay more attention to interlocutors when they assume they would need more help for comprehension. Attention would increase the depth of processing and encoding of linguistic information which would lead to greater alignment, and entrainment consequentially. In the current experiment, we replicate Branigan et al. (2011)'s experiment, by manipulating participants' beliefs about their interlocutor. One group of participants will play a picture matching/picture naming task with a virtual agent which is presented as highly competent (high competence condition), while another group will play the same task with the same virtual agent, presented, this time, as a beta version (low competence condition). Moreover, half of the participants will be asked to perform a secondary memory task (load condition). Additionally, all participants will be asked to perform a surprise follow-up memory task (word recognition task) that will allow us to investigate whether they attended to interlocutors' linguistic behaviour to a different extent in different conditions and whether attention boosts entrainment.