Language use is variable, such as we can call the same object umbrella or brolly. But what determines this variability? This thesis aimed at better understanding how speakers vary the words they use during dialogue, by investigating the mechanisms underlying lexical entrainment, or the tendency for a speaker to reuse a word that their partner has used before (e.g., using brolly after your partner used brolly; Brennan & Clark, 1996). Interestingly, speakers entrain to their partner’s word even when that means using a word that is not their speech community’s preference (e.g., using brolly instead of umbrella in British English; Branigan et al., 2011). There are two main accounts of how lexical entrainment works. Unmediated accounts explain it as the result of priming effects, in that a partner’s use of brolly made its lexical representation accessible from memory, thus enhancing its retrieval and reuse (Pickering & Garrod, 2004). In contrast, mediated accounts feature speakers’ beliefs in entrainment, with some such accounts suggesting that entrainment is aimed to enhance mutual comprehension (Clark, 1996) and others arguing that it is aimed to enhance our social affiliation (van Baaren et al., 2003). Critically, lexical retrieval, audience design, and social affiliation may vary both across situations and individuals, positioning lexical entrainment as a good candidate to inform how individual, interpersonal, and community-level influences affect language use. Thus, in four studies, we investigated individual, interpersonal, and community-level influences in lexical entrainment. The first research study examined the test-retest reliability of an online, webbased lexical entrainment task. Experimental items were objects with both a favoured name (umbrella) and a disfavoured but acceptable name (brolly), and we measured lexical entrainment as participants’ reuse of their partner’s disfavoured name (i.e., using brolly after their partner used brolly). Across two studies, we found that the task reliably elicited entrainment at the individual level both in the short- and the long-term, clearly suggesting that the tendency to reuse a partner’s lexical choice is stable within individuals and thus suggesting that individuals’ tendency to lexically entrain can be underlain by stable individual traits. The second study thus used this lexical entrainment task to test whether entrainment was predicted by schizotypy and age, which both correlate with individual differences in lexical retrieval, audience design, and social affiliation skills. Although entrainment was not predicted by schizotypy, it was positively predicted by age, suggesting that at least some mechanisms of lexical entrainment may undergo changes across the lifespan. The third research project investigated the causal relationship between social affiliation and lexical entrainment. Based on previous evidence that ostracism increases social affiliation as a compensatory strategy to recover social acceptance, we examined (i) the effects of ostracism on lexical entrainment, (ii) whether such effects were targeted to repairing a particular social relationship or increasing affiliation more generally, and (iii) whether ostracism effects were moderated by personality. Across two experiments, we found that lexical entrainment was positively affected by ostracism, but ostracised participants were as likely to entrain to a partner who had ostracised them as to a new partner; moreover, post-ostracism lexical entrainment was predicted by individual differences in neuroticism. Critically, these results suggest a social affiliation component to lexical entrainment. The fourth research project moved beyond looking only at lexical entrainment with a particular partner, to examine how speakers extrapolate community-level lexical knowledge from single linguistic encounters. Across three experiments, we found that participants generalised names across two partners depending on their community membership. These results suggest that social information is encoded during lexical processing, thus enabling the creation of community-level knowledge from single linguistic encounters. In sum, this research has important implications for understanding the variability of lexical choices. First, our individual differences findings indicate that lexical entrainment has potential to reveal what drives individual variation in how speakers make referential choices. Second, our group-comparisons findings indicate that the mechanisms of language processing are not encapsulated within the language system, thus suggesting that speakers vary their referential expressions based not only on linguistic processing but also on their beliefs and social dispositions. Taken together, these results suggest that the variability of language use is shaped by individual, interpersonal, and community-level influences, and by their interplay.