"long-term benefits to society" is one attitude. Surely you have better argument in defense of your field or humanities generally. Can you elaborate a bit on the kind of attitude you talk of?
I was preempting a response that I often hear when this debate comes up, which is that "the market" will fix the problem of underemployed PhDs. This often turns into an argument that research that has "practical value" (as judged by the general public) is somehow more important than that which does not. On Hacker News I sense that it's common to read "PhD" as "STEM PhD," and the response is "it's fine, they can just find positions in the industry."
This logic tends to fall short when applied to the humanities in general, and external (non-university) funding for humanities research is shockingly less than those in STEM fields. The National Endowment for the Humanities has an annual budget of $167 million, and the National Science Foundation, compared to $7 billion for the NSF (numbers from Wikipedia). Humanities research often tends to get mocked in debates on public funding: "Why are we, the taxpayers, paying this person to go look through archives in Germany looking for fragments of music by Robert Schumann?"
The humanities do not, in general, have "practical application," at least if practical application means monetary value (as it often does). As 1stop notes in a sibling, humanities research is often research for the sake of culture, or for the sake of research itself. I got into my field because I love it, but also because I think there is inherent value in knowledge. The number of people who will read my dissertation could likely all enjoy a meal together, but for those few people it matters deeply.
So there is some long-term benefit to society for humanities research (other people, likely also scholars, can benefit from it later), but it's not a benefit that's easily quantified. In today's society, this is a problem without a clear solution, and it's an attitude I see a lot on HN (with constant talk of valuation and whatnot).