Okay, it’s been a while, but at last here’s the second half from the earlier post about the use of that as a human-referring relative pronoun (HRRP). The issue before us is determining whether it’s all right to say something like:

(1) Everyone that knows me likes me

The point of contention is not whether this is too egotistical to be said, but rather whether the relative pronoun that is too demeaning to the humans composing everyone.  Are we required to revise (1) to

(1′) Everyone who knows me likes me ?

In the last post, historical usage revealed that it was historically common to use that as a relative pronoun with people.  (In fact, who wasn’t even a relative pronoun until the 15th century.)  However, history also showed that the behavior of relative pronouns is constantly changing.  So the question is a bit different in this post; we need to know whether the “people need who” rule is valid now, even if it wasn’t valid before.  And with that, we turn to the great repository of language that the world has ever known: the Internet.  I ran some quick Google searches, and here’re the results:*

X = who X = that X = whom
The people X I know 1700 438000 5990
The people X I saw 172 738 64
The man X I know 63 43400 43
The man X I saw 241 19600 18800
The people X know me 93700 36300 18
The people X saw me 2010 63 2

(The man X knows/saw me is omitted because of insufficient attestations.)  The high-level summary is that sometimes who is preferred to that, and sometimes it’s the other way around. On occasion, whom asserts itself as well, although it’s never the most popular form.  Okay, that’s great!  Now we know that that is an acceptable HRRP, just as it has been throughout history (see previous post).  So, prescriptivists, would you mind terribly dropping the claim that who is for people and that is not?  Much obliged.

But what’s more interesting is that there is a clear pattern to the usages. Note that that is most common when the relative clause contains a subject (I) but no object, and least common when the relative clause contains an object (me) but no subject. Who runs the other way, appearing mostly with the know/saw me clauses. If you’ll permit a bit of terminology, this shows people prefer that in Object-Extracted Relative Clauses (ORCs) and prefer who in Subject-Extracted Relative Clauses (SRCs).

(A relative clause can be thought of as a sentence turned inside out; one noun phrase is moved from its position inside the sentence to a position of prominence before everything else. If the subject is extracted, you get something like The man ate the fish -> The man that ___ ate the fish. If the object is extracted, you get something like The man ate the fish -> The fish that the man ate ___. The former is an SRC, and the latter is an ORC.)

This same result, the SRC-ORC distinction, pops up for different verbs (I tested pass as well) and different pronouns (they know/know them worked too). (Unfortunately, I couldn’t test to see if longer NPs in the relative clauses worked the same due to the limitations of online searches, but I’m willing to bet that the same is true for non-pronominal or long NPs.)

What we’re seeing here is that both who and that are acceptable as HRRPs, despite what prescriptivists say. But the interesting thing is that different contexts prefer one or the other.  I’ve got a conjecture about this: who is preferred in simpler contexts, and that in more complicated ones. In a psycholinguistic sense, it’s plausible that who is more complex than that, because that is the default relative pronoun. You have to check when you use who that you’re referring to a person (or other sentient being), but you don’t have to do that with that. When you’re working with a more cognitively taxing context, it’s costly to expend still more effort to use who than to settle for the default that.

There’s been a ton of psycholinguistic research that shows that ORCs are harder to produce and comprehend than SRCs are, so that might explain the differential deployment of who and that. I don’t know. But following on Florian Jaeger and Roger Levy’s work on that-omission being tied to processing and production difficulty, this strikes me as a potentially interesting conjecture. Of course, testing that would require corpus annotation, a controlled study, and all that jazz that I decided to take a break from after my comps paper. So it goes.

Summary: That is a perfectly fine relative pronoun, even for people. In fact, in many contexts, that is more common as a human-referring relative pronoun than who. (“The people that I know”, for instance, is more common than “the people who(m) I know”.) Interestingly, that seems to be more common for more difficult relative clauses.

*: Google uses heuristics to guess how many webpages use a string, so the estimated number of results shown on the first page is often inflated. By clicking a few pages of results down (I went to page 10), you get a substantially more accurate estimate. For instance, Google claimed there were “about 3,090” hits for the man whom I know at first, but when I attempted to access its tenth page of results, it recanted and claimed that there were actually only 43 results. WOW. So all of the numbers reported above are from the tenth, or last, page of results. Even then, these should be treated as highly variable estimates; a difference should be at least an order or two of magnitude before it is trusted.