I had a dream: Subjunctives in English and elsewhere

I’d been dreaming that I was sleeping really well.  In fact, I wasn’t sleeping well at all.

So, I laid in bed staring at the ceiling for what felt like an interminable amount of time.  Finally got up and looked at the clock: 4 AM. Before I woke up, I’d been dreaming that I was sleeping really well.  In fact, I wasn’t sleeping well at all.

Before I woke up, I’d been dreaming that I was sleeping really well.  In fact, I wasn’t sleeping well at all: English is my native language, but I’m not sure that what I just said makes sense.  It seems hopelessly unclear.  Is it the case that “in fact” I wasn’t sleeping well in the dream, or that “in fact” I wasn’t sleeping well in real life?  In fact, I wasn’t sleeping well in real life–I was just dreaming that I was.  I don’t know of a way to disambiguate that in English.  What is called for here: a language with a robust past tense of the subjunctive.


The subjunctive mood is the term that is usually given for grammatical structures that express things that are in the realm of wishes, desires, opinions, and possibilities, as opposed to things that are facts.  It just barely exists in English, and as far as I know, in English it is always optional.  To the best of my knowledge, the subjunctive only exists for the verb to be.  Here’s what it looks like, in typical American English and in the Pacific Northwest dialect.  This is a way that you can give someone advice:

  • Typical American: If I was you, I wouldn’t do that.
  • Pacific Northwest: If I were you, I wouldn’t do that.

The difference: in typical American English, you would use the past tense form for the first person singular: was.  In the Pacific Northwest, you use were.  We use was for the past tense, of course–it’s only in the subjunctive that you see this weird use of the were form.  You use it for other persons, too, in the subjunctive:

  • Typical American: If he was smarter, he wouldn’t have done that.
  • Pacific Northwest: If he were smarter, he wouldn’t have done that.

Well: English does not have a robust past subjunctive at all.  Some languages do, though.  How might I talk about my dream in one of them?  Let’s consider some options.

Modern colloquial French does not have a robust past subjunctive at all.  Literary French does, though–a leftover from earlier forms of the language, and what we would be looking at here is an ongoing action, so it would be the subjunctive imperfect that we’d be using.  (I think–again, I’m not a native speaker.)  Here’s an attempt at both of them, neither of which I speak natively, or even well:

Modern colloquial French: Je rêvais que je dormais bien, tandis que de fait je ne dormais point bien.

Literary French: Je rêvais que je dormisse bien, tandis que de fait je ne dormait point bien.

In contrast with modern colloquial French, modern colloquial Spanish does, in fact, have a robust past subjunctive.  “Robust” in the sense that people do actually use it.  Let’s try that:

Soñaba que durmiera bien, aunque de hecho no dormía nada de bien.

…aaaaaand, with that I see that in Literary French and in modern colloquial Spanish, you can express the case where in real life I wasn’t sleeping well at all, but I don’t see a good way in either language to convey the situation where it’s in the dream that I wasn’t sleeping well.  Have I fucked up all four languages (English, modern colloquial French, literary French, and modern colloquial Spanish) here?  Forgive me, ’cause it’s not even 5 AM, and I didn’t sleep well last night.

Scroll down past the video of the somewhat cute song L’Imparfait du subjonctif, “The Imperfect Subjunctive” (Pourtant je le pus et vous pûtes, hee hee hee) if you want to read the English notes.  Otherwise: go back to bed.


English notes

To disambiguate: To differentiate between two possible senses (meanings) of something (“of an utterance,” as a linguist would put it).  In computational linguistics, it usually means to find the intended sense.

  • In the French sentence L’étagère plie sous les livres (‘The shelf is bending under [the weight of] the books’), it is necessary to disambiguate the sense of livres (which can mean ‘books’ or ‘pounds’ and is masculine in the former sense, feminine in the latter) to properly tag it as a masculine noun. (Ide, Nancy, and Jean Véronis. “Introduction to the special issue on word sense disambiguation: the state of the art.” Computational Linguistics 24.1 (1998): 1-40.)
  • Lapata and Brew (1999) and others have shown that the different syntactic subcategorization frames of a verb such as serve can be used to help disambiguate a particular instance of the word. (Gildea, Daniel, and Daniel Jurafsky. “Automatic labeling of semantic roles.” Computational Linguistics 28.3 (2002): 245-288.)
  • When you search for information regarding a particular person on the web, a search engine returns many pages. Some of these pages may be for people with the same name. How can we disambiguate these different people with the same name? (Bollegala, Danushka, Yutaka Matsuo, and Mitsuru Ishizuka. “Extracting key phrases to disambiguate personal names on the web.” International Conference on Intelligent Text Processing and Computational Linguistics. Springer, Berlin, Heidelberg, 2006.)

For example: I giggled about the lyrics Pourtant je le pus et vous pûtes because when spoken, it is ambiguous: it could mean either however, I could and you could (the intended sense) or however, I stink of it and you whore.  In the latter sense–which, I will note, makes no sense, and we will return to that fact momentarily–it would be written pourtant je le pue et vous pute.  So, it’s not ambiguous in writing, but it is à l’oral. 

Now: almost everything that you will say, hear, write, or read today will be ambiguous in some way.  But, humans are so good at disambiguating that we notice that ambiguity only rarely.  How do we do it?  It’s mostly mysterious, but our behavior is consistent with the notion that we calculate the set of possible meanings and select the one which is most probable.  That’s a very different thing from our normal way of thinking consciously about this, in which I might say that “I stink of it and you whore” makes no sense.  “Makes no sense” implies that there is a binary distinction–either something “makes sense,” or it doesn’t.  When you talk in terms of probabilities, then you are thinking of meanings as something that can be more or less, which is very different from being, or not.

How do computer programs do this?  Computational linguists build systems that work more or less the way that we think humans work: determine the set of possible meanings, calculate a probability for each one, and select the most-probable of the set.  What happens if there’s a tie? Well…read this paper by Antske Fokkens.

Screen Shot 2020-07-09 at 5.33.57 AM

What computational linguists do during global pandemics

It has been a real satisfaction to see the flow of sometimes quite good papers about how our field can be brought to bear against the current global plague. 

Dear Mandelbrot,

We’re writing to request that you review 1-5 papers on a 3-day turnaround for the Association for Computational Linguistics’s workshop on natural language processing for COVID-19.  This is the Association for Computational Linguistics’s first “emergency workshop.”  As such, we have been experimenting with accepting rolling admissions.  We had a very steady stream up to the deadline–definitely enough to keep our pool of reviewers (and chairs) busy.  Then, at the submission deadline, we got completely inundated with new submissions, and now we are trying to deal with that flood of papers to be reviewed.  Can you help?

We are aware that reviewing on a 3-day turnaround is a big ask.  It has been a real satisfaction to see the flow of sometimes quite good papers about how our field can be brought to bear against the current global plague.  We hope that you will get the same satisfaction from joining us from this all-hands-on-deck call for reviewers.  Can we count on you to join the crew?

Best wishes, 

Zipf and Karin, for the organizing committee


The good news: it’s satisfying to look at something that you wrote in a panic at 2 AM in your native language and see that there is some sort of coherence to it.  If you spend as much time as I do trying just to spell correctly in a language that’s not natively yours, it is a real relief to have stuff just flow sometimes.

The bad news: I can’t see any connection between the aquatic lexical field here and the topic of recruiting reviewers.  The lexical field brings into focus the weight of the recruiting burden that the workshop has generated, but other than that, I can’t see any real rhetorical relation.