Is it worth it to go to grad school in computational linguistics?

I received an email from Julian Wilkinson, an undergraduate double-majoring in Linguistics & CS (name used by permission) who is trying to decide whether to go to grad school or go straight into industry. The questions were thoughtful and as I typed out my response, I thought that perhaps other people might have similar questions and that putting my answers out in a blog post might be helpful.

Here, with permission, are Julian’s questions and my answers:

Julian: While doing research on computational linguistics programs and their faculty I stumbled upon your LinguistList profile, where, when I read it, I found a part of your undergraduate experience very relatable. In regards to your decision to major in linguistics at Cal, you said, “It took me the rest of the semester … to convince myself that I could major in something I perceived as very impractical.” I’ve personally struggled with this feeling, but I’m really interested to know how you overcame it.

Also, I find myself at a proverbial fork in the road, where I feel like I’m in a position to choose to either go to graduate school, where I’d now focus on preparing myself to be a stronger candidate for a computational linguistics program, and entering the industry, where, instead, I’d focus on finding internships and do projects to bolster my resume. If you have the time and don’t mind, I have a few questions about graduate school and computational linguistics in general I’d like to ask you:

Graduate School Questions:

1. How did you end up convincing yourself to major in linguistics?

2. Why did you choose to go to graduate school?

3. Are there any reasons you would discourage someone from attending graduate school?

Supplementary, Computational Linguistics Questions:

4. What books or resources would recommend to someone new to computational linguistics?

5. Do linguistic rules and analysis still play a large role in computational linguistics today?

6. Are there any sociolinguistic applications of computational linguistics?

Thank you for taking the time to read this email. I hope to hear back from you.

Emily: Thank you for your query! I will try to give some helpful answers.

1. How did you end up convincing yourself to major in linguistics?

I decided I loved it too much not to, and that I could surely make something of it. Things are very very different now — computational linguistics in particular is a very lucrative field. A double major in Ling + CS sets you up very nicely.
 
2. Why did you choose to go to graduate school?

That was easy. I *love* school — and when I learned that there was a path that would allow me to stay, it was the obvious choice for me.[1] (I attended grad school with the goal of becoming a professor; succeeding in that entails a lot of luck and I was lucky. I also gave myself a deadline of 5 years post PhD to find a tenure track job, not being willing to do the contingent faculty thing long-term.)
 
3. Are there any reasons you would discourage someone from attending graduate school?

Yes — but the answer is very different for Master’s (MA, MS) v. PhD. For an MA/MS, it’s a question of costs & benefits. How much will the program cost? How important is the additional training for what you want to do? I’d say that for doing compling in industry, the MS can be a big value add, even if you are doing Ling + CS in undergrad. Also, many Master’s programs can serve as a relatively low-risk way to find out if you like grad school & want to continue.

For the PhD, I always ask prospective students “What do you want a PhD for?” There are pretty big opportunity costs to getting a PhD (you could be making a lot of money during those 5 years, if you just went to industry), doing a PhD can be stressful, and if people are doing it because they want to be faculty, it’s important to understand that faculty positions are few and far between. On top of that, grad school can be very stressful. So there has to be a good reason to want to do it, and “I absolutely want to be a prof, no other career path will work for me” won’t cut it. Nor will “I can’t think of anything else to do right now.”

4. What books or resources would recommend to someone new to computational linguistics?

The classics are Jurafsky & Martin’s text book, the NLTK book by Bird et al, and Manning & Schutze’s text book. Also check out the #NLProc tag on twitter. For someone coming from CS only, I also recommend my book:

Bender, Emily M. 2013. Linguistic Fundamentals for Natural Language Processing: 100 Essentials from Morphology and Syntax. Synthesis Lectures on Human Language Technologies #20. Morgan & Claypool Publishers.

Note that if your institution has a subscription, you should be able to get an e-copy through your library for free. This might still be helpful for someone with a linguistics degree, but less so (you probably know much of it).
 
5. Do linguistic rules and analysis still play a large role in computational linguistics today?

Yes & no. The conferences are dominated by machine learning (with deep learning/neural nets currently in vogue), but one finds both kinds of work in industry in terms of systems development. In terms of evaluation, error analysis, resource development (think WordNet, FrameNet or treebanks like the Universal Dependencies treebanks or Redwoods), and feature or system architecture design there is large scope for linguistic insight, often untapped. We need more people in the field with an understanding of linguistics and how it applies!
 
6. Are there any sociolinguistic applications of computational linguistics?

Yes! There are a wide range of possibilities here, including looking at how people use language to present themselves (e.g., Marin et al 2011, Morgan et al 2013) or indicate their stance in dialogue (e.g. Freeman et al 2015), to looking at how existing technologies work differentially well depending on the demographic characteristics of the user (e.g., Jørgensen et al 2015, Tatman 2017).

Footnote:

[1] I should acknowledge another way in which I was lucky. I finished my undergrad degree without any student loan (or other debt), thanks to support from my parents, and wasn’t supporting anyone else at the time, so it was (almost) feasible to live on grad student fellowships/TAships while I did my PhD. I say almost because the Bay Area in the late 90s was already pretty expensive. A part-time job translating error messages for a start up made up the difference. In sum, this made the decision to pursue a graduate degree easier for me than it might have been for many of my peers.