In this paper, I give an account of constructions expressing similarity such as 'like John' and 'like a lawyer'. The main point of the paper is that in 'like a lawyer', the indefinite receives a generic interpretation, which explains why under its most available reading, 'John looks like a lawyer' is equivalent to 'John looks like a typical lawyer'.
However, this indefinite is generic in a surprising way. Generic quantification is standardly thought to be brought about by a silent quantificational adverb, Gen , bearing a meaning akin to 'generally'(cf. Krifka et al., 1995). It is therefore expected, on the standard picture, that an indefinite that can receive generic interpretations should also be bound by explicit quantificational adverbs, as for instance in 'a bird flies' ≈ 'typical birds fly', parallel to 'a bird rarely flies'≈ 'few birds fly'. However, indefinites embedded by 'like' escape this generalization: 'John looks like a lawyer'≈'John looks like a typical lawyer', but 'John rarely looks like a lawyer' ≠ 'John looks like few lawyers'.
To solve this puzzle, I propose that 'like' comes with a generic quantifier that is lexically hard-wired in its lexical entry, and show how this makes a number of surprising predictions which all turn out to be correct; I also discuss implications for theories of genericity.
Along the way, I also analyze properties of 'like' that are not necessarily linked to genericity, mainly: (i) it is a gradable expression over a closed scale, since it supports proportional modification such as in 'the DNA of humans is 99% like that of chimps'. (ii) It can be modified both by scalar modifiers like 'much' and by 'with respect to' phrases like 'with respect to size', in similar but non-identical ways. (iii) It gives rise to homogeneity (cf. Križ, 2015 a.o.), as 'John is like Mary' suggests they share all relevant properties, while 'John isn't like Mary' suggests they share none of them.