We can see that plagiarism is a funny concept in part from the fact that what counts as plagiarism in one field will not necessarily count as plagiarism in another. A fairly straightforward example is in contract law: contract lawyers borrow language from each other all the time in ways that would certainly count as plagiarism in some other fields. But this, of course, is because nothing hangs on this in particular. In general, lawyer reputations are not made by designing completely original contracts; a society in which they were would certainly be contractually creative, but would also be quite bizarre, since such a society could not possibly have effectiveness of contracts as its central idea in contract law. If everyone's primary interest is simply contracts that do what they are supposed to do, copying contract language that works is obviously the way to go. We could imagine a system of contract law in which lawyers would be fined if they produced contracts using contract language whose source was not cited, but it's hard to see what this would accomplish beyond increasing the cost of hiring contract lawyers. We see this quite widely. Journalistic citation practices would not in general be kosher in academic contexts; there are entire fields in which books and speeches are ghostwritten and this is taken for granted (when was the last time that you heard about a politician plagiarizing his speechwriters by failure to cite them?), but ghostwriting is arguably the single most serious plagiaristic offense of which a student can be guilty; almost everyone allows 'common knowledge' exceptions to plagiarism rules, but what counts as common knowledge varies from field to field; and, perhaps most noteworthy of all (because often forgotten), virtually all academics, who are almost all intensely anti-plagiaristic in one way, regularly do things in the classroom that almost certainly would be regarded as plagiaristic elsewhere and sometimes even in their own fields ('common knowledge' is never so expansive as it is in the classroom).
Following Posner, Clark gives the following characterization of genuine plagiarism:
An act of plagiarism requires 1) the substantial copying without credit of one person’s language by another; 2) that this copying is done fraudulently, that is, to fool the reader into thinking that the work is original; 3) that the plagiarist acts with the clear intention to trick the reader; 4) that the act has potential negative consequences for the reader and the original author.
This conceives plagiarism along the lines of fraud: the plagiarist on this understanding presents themselves fraudulently to the detriment of others. (One might wonder why the distinction between (2) and (3); it's possible that this is an accidental splitting, but the two can be interpreted differently, with (2) identifying form and (3) identifying goal or end.) I think analogizing plagiarism to fraud is as problematic as analogizing it to theft, but the condition (4) is particularly important, I think, because (as I've argued before) something like it is the only thing that allows our practices with regard to plagiarism to make sense at all.
Plagiarism is a problem (one might equally say that copying is only plagiarism) only in fields where it distorts essential metrics in ways that are recognizably detrimental to the field. In practice there are three metrics whose distortions can be so detrimental: money, various kinds of reputation metrics, and grades (although grades can be considered a reputational metric themselves). Fields like academia and journalism put a focus on plagiarism because these fields are heavily reputation-dependent; fields like publishing need anti-plagiarism rules in place to protect income; and schools come down hard on plagiarism because they need to maintain the reputation of their degrees, certifications, and the like. There aren't many other situations in which copying would even be an issue, beyond those involving distortion of the field's ways of distributing profit and reputation.
Rather than thinking of plagiarism as a one-size-fits-all category, then, we need to think in terms of cost and protection of incentive in the field at large, whatever the field may be. This is the real reason for the variations in what counts as plagiarism. The money and reputations of folk singers don't depend on their being the only people who can sing a given song in the way it does in the standard music industry; a folk singer delivers performance rather than song, and thus folk singers generally are better off if they are able to share songs freely. When people listen to folk singers -- or any singers other than those who have a standard playlist, for that matter -- they don't do so in order to hear originality but in order to hear quality. Being the originator of a song can influence a folk singer's reputation, and hence money, positively, but it does so in a way that usually requires a wider diffusion of their song than they themselves can usually manage. (This is, to go on a brief only-partly-relevant tangent, the reason for a divide that is often not acknowledged in major music and print publishing industries, which is that while authors/songwriters who can guarantee wide distribution benefit from very strict copyright protections, authors/songwriters who cannot guarantee such wide distribution often benefit from their works being copied without the complications of permissions. The more people who copy such an originator, the greater the probability that monetary channels will start paying off, and the greater the probability that the work will endure a long time as a source of money and/or reputation. What these smaller authors and songwriters really require is not anti-copying rules but attribution rules -- they need to keep the connection to their name intact. Copyright laws, however, are generally imitation-focused rather than attribution-focused, because large-scale publishing concerns don't have to worry much about attribution, but about whether they have any competitors, whether it be another publishing house or someone with a copy machine or scanner. This is part of the reason why the public is constantly irritated by copyright laws: people take attribution to be far more important, and find it to be far more intuitive, than technical rules about what you can and can't copy with your own time and money.)
In short, what we should be doing in talking about plagiarism is talking about why the activity in question distorts measures that are important for monetary or reputational reasons. Indeed, I'm often attracted to the view that we should stop talking about plagiarism at all, and instead just talk about the problems different activities cause for the field. Another alternative would be the one Clark takes, which is to identify a particular problematic activity as plagiarism in the proper sense, or as the key kind of plagiarism, and sort out every other kind of activity according to its relation to this key one. Regardless, we do need to think critically about what we say when discussing plagiarism.