This all sounds Turkish to me


We all make mistakes. This is true of haircuts, job interviews, extra slices of cake and crucially, language. So here is one type of language error that you may have heard before but never thought much about. How does this sentence sound to you?

(1 )

The key to the cabinets are on the table.

Some people say that it sounds fine. Others say that it does not sound quite right, but that it is definitely better than (2):

(2 )

The key to the cabinet are on the table.

Everyone agrees about (2): it sounds terrible, no one would say that! But if you think about it, from a purely grammatical viewpoint both sentences are equally WRONG. In both cases, the verb "are" is incorrect because it should agree with "key" : it is the key that is on the table, not the cabinets! But for some reason, people like the sentence more when the verb agrees in number with "cabinets", which is a modifier (not the head) of the subject phrase. So why does (1) sound better than (2)?

Some examples of English agreement attraction errors.

Errors like (1) are called agreement attraction errors and they occur quite often in English. And "cabinets", the noun that the verb wrongly agrees with, is called an attractor. This name conveys the intuition that "cabinets" deceitfully lures the verb to agree with it in number instead of with the real controller, "key".

What we don't know though, is why attraction happens. What misleads people into making these errors? You might be thinking "Jeez, it's obvious: "cabinets" is closer to the verb than "key"! Maybe people just produce a verb that agrees with whatever noun is closer to it". Ok, you have a point, but notice that these errors occur even when the attractor isn't next to the verb. Look at the first example on your left: the noun "errors" triggers attraction despite the fact that the real controller, "classification software", is actually closer to the verb. So, although linear proximity might increase attraction, it cannot be its only cause.

How common are these errors? Well, we knew since the 90's that people made agreement attraction errors in speech. But we didn't know whether they also noticed them in comprehension. That is, do people fail to realize that there is an agreement error in a sentence like "The key to the cabinets are?"

In 2009, a group of researchers from the University of Maryland, Matt Wagers, Ellen Lau and Colin Phillips set out to answer this question. They brought a bunch of students to their lab and presented them with sentences like (3) below. They varied whether the sentences were grammatical or ungrammatical ("was" vs. "were"), and crucially, whether they contained a singular or plural attractor noun ("cabinet" vs. "cabinets"). Each sentence had 4 different versions, but of course, each participant in their experiment only saw one:


The key to the cabinet was on the table.


The key to the cabinets was on the table.


The key to the cabinet were on the table.


The key to the cabinets were on the table.

The participants read the sentences and were asked to judge whether the sentences were acceptable or unacceptable. The researchers defined ACCEPTABLE SENTENCES as sentences that sounded like normal English sentences, and could be said by any native English speaker. Conversely, UNACCEPTABLE SENTENCES were defined as sentences that an English speaker would be unlikely to say, because they sounded weird or unnatural.

The results from Wagers and colleagues (2009). The participants in their experiment accepted ungrammatical sentences more often when they contained the attractor "cabinets" (3d).

The plot shows their results. The x-axis shows the 4 experimental conditions. The y-axis indicates the proportion of cases where participants judged sentences as acceptable. In the grammatical conditions, participants rated the sentences as acceptable most of the time (above 90%). In these conditions, it didn't seem to matter whether the sentences contained a singular or plural attractor.

In contrast, the ungrammatical sentences showed an interesting pattern. Whereas sentences lacking an attractor like (3c) were deemed unacceptable most of the time, sentences with a plural attractor like (4d) were accepted... more than 50% of the time! This showed that these sentences were perceived as more acceptable due to having the attractor "cabinets" inside them. As a result, Wagers and colleagues concluded that indeed, English speakers make attraction errors and can be mislead by plural nouns and fail to notice ungrammatical sentences.

So now we know that English speakers are susceptible to attraction errors in comprehension. Fine, no one is perfect. But why do these errors happen?

Here is a VERY SIMPLE EXPLANATION: people fail to notice attraction errors in comprehension precisely because they often make them in production. I mean, it would make sense, no? People make these errors when they speak. So they probably hear these errors often and are more accustomed to them which results in their noticing them less in comprehension.

It's a simple explanation, and it seemed obvious for English. But this also meant that we couldn't figure out how to test it scientifically!

This is where Turkish comes in. Because Turkish has one agreement property that makes it different from languages like English or in fact, from any of the languages tested to-date: in Turkish, when a subject is plural like "The detectives" (Dedektifler), Turkish people prefer singular (e.g. "works", çalıştı) instead of plural verbs (e.g. "work", çalıştılar). No one knows exactly why, but one possibility is that since Turkish marks plural number by adding the suffixes -lar/-ler to both nouns and verbs, people might avoid plural verbs with plural subjects to not repeat the plural marker twice: e.g. "The detectives work" ("Dedektifler çalıştılar", dispreferred) vs. "The detectives works" ("Dedektifler çalıştı", preferred).

Regardless of the specific reason, it is known that Turkish speakers avoid producing sentences with plural verbs and plural subjects. This creates a language where, in contrast with English, people should not hear this combination often in speech. So, we wondered: do Turkish speakers show agreement attraction in comprehension? Or is agreement attraction absent in Turkish, in contrast with English?

My Ankara trip. In the pic you see me, Bilal and Orhan, from the Middle East Technical University.

To answer this question, I flew to Ankara and I joined a group of researchers at the Middle East Technical University: Martina, Bilal, Duygu and Orhan.

First, we decided to verify that Turkish people really preferred plural to singular verbs with plural subjects. We constructed sentences like (4) in different versions. We had singular subjects with singular verbs (4a) and plural subjects with plural verbs (4b). But also, we designed two versions where the subject and verb mismatched in number: either the subject was singular and the verb was plural (4c) or the other way around (4d).

We were especially interested in cases where the subject was plural and the verb was singular (4d). These cases would be totally bad in English but they should be ok in Turkish, since Turkish speakers might prefer singular verbs to avoid repetition of the plural marker. And in fact, this is what we saw: Turkish speakers judged sentences like (4d) as acceptable almost 99% of the time!

Turkish subject-verb agreement preferences. In contrast with English, Turkish speakers accept sentences with plural subjects and singular verbs (4d) almost 100% of the time.


The detective works.

Dedektif çalıştı.


The detectives work.

Dedektifler çalıştılar.


The detective work.

Dedektif çalıştılar.


The detectives works.

Dedektifler çalıştı.

But what really surprised us is what we saw next. To check whether Turkish speakers were sensitive to attraction, we tested sentences similar to those used in English before. We couldn't use sentences exactly like "They key to the cabinets" in Turkish. So we used sentences where the attractors were possessor nouns like "students" in "the students' advisor". We wanted to see whether the possessors would mislead Turkish speakers to accept ungrammatical sentences, just like "cabinets" had done to English speakers. We predicted that if people made these errors in comprehension due to hearing them often, then Turkish speakers should be immune to them, as they are not likely to hear plural subject-plural verb combinations in their language.

turkish attraction
Agreement attraction in Turkish comprehension. Our participants accepted ungrammatical sentences more often when they contained a plural attractor (5d).

Surprisingly, this is NOT what happened: in fact, our Turkish participants showed a robust attraction effect! They accepted ungrammatical sentences with plural attractors around 25% of the time (5d), almost 10% more than the fully ungrammatical sentences, which did not have any attractor (5c).


The student’s advisor suddenly faints.

Öğrencinin danışmanı birden bayıldı.


The students’ advisor suddenly faints.

Öğrencilerin danışmanı birden bayıldı.


The student’s advisor suddenly faint.

Öğrencinin danışmanı birden bayıldılar.


The students’ advisor suddenly faint.

Öğrencilerin danışmanı birden bayıldılar.

So this is where we are now. It is kind of puzzling: despite disprefering plural subjects with plural verbs, Turkish speakers are still susceptible to attraction errors. To us, this means that attraction errors in comprehension do not occur simply because people hear these errors often in production (although admittedly, we haven't tested whether Turkish people produce attraction errors yet... If we are on the right track, they shouldn't!). More generally, our results suggest that attraction errors in production and in comprehension might not obey the same principles, which is an intriguing possibility. But then, what causes attraction in comprehension? This, I think, will be a harder question to answer. Stay tuned for updates!

My last dinner in Ankara. After a week of running Turkish participants, we felt that we deserved a drink.