Grammatical accuracy (3 hours)

↵ Back to module homepage

This last one is pretty complicated.

In the previous sections we have seen evidence that the parser is incremental and that it is interactive. The last question is whether it is grammatically accurate. To understand this question, we first need to learn about a new kind of experiment and a new kind of effect.

The filled gap effect

In the previous sections we talked about the garden path effect. There is another kind of experiment looking at something called the filled gap effect. It's a bit similar: it's also a way to see what happens when people misinterpret a sentence and then have to change their interpretation. But the details are a little bit different.

To understand the filled gap effect, you need to understand that words sometimes move in a sentence. Think about the way we make a question in English, such as What are you laughing at? We can make a question sentence in English via several steps:

  1. Starting with a declarative (non-question) sentence, such as: You are laughing at her sillystory.
  2. Find the thing that the question will be about, and replace it with what. Now, the sentence "You are laughing at her silly story" will turn into "You are laughing at what?"
  3. Move the "what" to the beginning of the sentence. Now, "You are laughing at what?" will change into "What are you laughing at ____?"

So the way we make sentences in English [usually] is we move a question word to the front of a sentence. (This description is a little bit oversimplified, but it's enough for our purposes right now.)

When words move like this in a sentence, there are two key parts: the gap and the filler. When a word moves, it leaves a "gap" (see the blank space I left in sentence #3 above). I.e., the space that the word moved away from. We don't normally write the gap when we write, and we don't normally pronounce it when we speak. But in the abstract language structure in our mind, it is supposedly there (according to linguistic theory). The filler is the thing that moved out of the gap (e.g., the word what).

When we read a sentence and we see a filler (a word that has been moved, such as what), we might start looking for the gap (the place that it moved from). We need to do this to understand the sentence (i.e., when we read What are you laughing at, we need to understand that "what" is actually the object of "laughing at").

You can imagine how this works using a metaphor/analogy. Imagine you're walking in a beautiful garden and you find a flower that has been ripped out of the ground and is lying on the pavement. You don't want the flower to die, you want to put it back in the ground so it can survive. The bottom of the flower has a big clump of dirt, where its roots have been ripped out of the ground. So you know that somewhere in the garden there must be a hole in the ground, where this flower was torn out of. So you might pick up the flower and look for the hole in the garden so you know where to put the flower back. In fact, this is also what we do when we read a sentence with words that have moved: when we see the word what we might recognize "hey, that word doesn't belong here!", and in order to interpret the sentence we mentally search for the "gap" in the sentence where the word what came from.

Imagine you read a sentence The teacher asked what you laughed about. The gap is at the end of the sentence ("The teacher asked what you laughed about ____"; the sentence was probably originally formed from something like "You laughed about the joke", which turned into "You laughed about what", then moved to the front to make "what you laughed about _______", and then finally the full sentence "The teacher asked what you laughed about _____".) Let's think step-by-step about what happens in a reader's mind when they read this sentence:

  1. They read The.
  2. They read teacher. They probably recognize that The and teacher can go together to make a phrase (The teacher). They probably also guess that this is probably the subject of the sentence.
  3. They read asked. They probably recognize that this is a verb, and they start to wonder what the teacher asked.
  4. They read what. ALERT, ALERT! The word what is a filler! That means it's moved from somewhere else in the sentence! The reader now realizes that they need to find the gap: they need to find where the what came from! From now on, as they continue reading the sentence, they will need to check at every word to see if that location could be the gap where what came from!
  5. Next they read you laughed. Could you put a gap here? "The teacher asked what you laughed ______." Does that work? That would mean the question was formed from a sentence like "You laughed what?" or "You laughed the joke." But those sentences aren't grammatical in English; in English we laugh at a joke, we don't "laugh a joke". (of course in Chinese this would be different.) Therefore, this must not be the gap where what came from. Got to keep searching!
  6. Next they read at. Could you put a gap here? "The teacher asked what you laughed at _____." That works perfectly! "You laughed at what?" or "You laughed at the joke" are fine in English. Hooray, now we found the gap where what came from.

But now let's imagine a slightly trickier sentence:

Where is the gap here?

It's actually like this:

In other words, the question wasn't formed from a sentence "The team laughed about what?" We know what they laughed about: they laughed about Greg dropping something. But we don't know what Greg dropped. So the underlying sentence is: "The team laughed about Greg dropping what?"

This might cause some special confusion to people as they are reading. Again, let's look step-by-step at how people might read this sentence.

  1. They read The.
  2. They read teacher. They probably recognize that The and teacher can go together to make a phrase (The teacher). They probably also guess that this is probably the subject of the sentence.
  3. They read asked. They probably recognize that this is a verb, and they start to wonder what the teacher asked.
  4. They read what. ALERT, ALERT! The word what is a filler! That means it's moved from somewhere else in the sentence! The reader now realizes that they need to find the gap: they need to find where the what came from! From now on, as they continue reading the sentence, they will need to check at every word to see if that location could be the gap where what came from!
  5. Next they read the team laughed. Could you put a gap here? "The teacher asked what the team laughed ______." Just like before, that doesn't work; there cannot be a gap here. Got to keep searching!
  6. Next they read about. Could you put a gap here? "The teacher asked what the team laughed about _____." That works perfectly! "The team laughed about what?" or "The team laughed about the joke" are fine in English. Hooray! Now the reader thinks the sentence is over; they think they know where the what came from.
  7. Next they read Greg. Hey, that doesn't make sense! They already thought the sentence was done (basically, "What did the team laugh about ____?"). But now there's an extra word: "What did the team laugh about Greg". It's not grammatical! It's not what the reader expected! It doesn't make sense! Need to think a little more and figure out what this sentence means, and perhaps revise the way I had interpreted it before.

Eventually the reader will (hopefully) read on further and understand what the sentence means. But at the moment they read Greg, they may be surprised and confused, and that may make them slow down.

The reason they get confused when reading Greg is because the place they expected a gap to appear was actually FILLED. A gap is a blank space in the sentence structure; they expected the sentence to be finished here. But where they expected to read nothing, they instead read the unexpected word Greg. The "gap" was filled by another word. That's why we call this kind of experiment the filled gap effect.

As always, to measure an effect, we need to compare it against something. (Remember when we examined the garden path effect, we couldn't just see if people read was slowly; we had to see if they read was in the reduced sentence more slowly than they read it in the full sentence. And to examine the priming effect we couldn't just see if people react to EDIT-BEE slowly; we have to see if they react to EDIT-BEE more slowly than they react to BUTTERFLY-BEE.) To do this, we usually compare a sentence that is very similar to that one but which doesn't involve movement (and thus doesn't involve any gap):

You don't need to worry about learning all the detailed syntax theory. The important thing is, what is a word that has moved from somewhere else in the sentence, whereas if is not. And "The teacher asked what the team laughed about..." sounds like it could be a complete sentence, so people will be surprised when they see Greg; on the other hand, "The teacher asked if the team laughed about..." is not a full sentence, so people should know there must be more words coming, and they won't be surprised when they see Greg. Therefore, the bottom line is that we expect Greg to be read more slowly in the WHAT sentence than it is read in the IF sentence. If that happens, that will be a filled gap effect.

To summarize, here's the typical design of a filled gap experiment:

In the WH clause, "Greg" should be surprising and confusing (and thus read slowly) because it's filling the place where the reader expects a gap. On the other hand, in the IF clause, "Greg" is not surprising or confusing. Thus, we expect that "Greg" will be read more slowly in the WH clause than it is read in the IF clause.

This is in fact what happens; a famous example is the experiment by Stowe (1986), which we will discuss more below.

Islands

That stuff above was already pretty complicated, but language can get even more complicated than that! That stuff was just background information; we haven't even gotten to the real topic of this section yet.

As we discussed above, words sometimes more around in a sentence, such as when we create a question in English. But that's not the whole story. In fact, words can't move out of any part of a sentence. Some sentences have islands: places that no word can move out of. (Do you remember the American TV show Lost from about 15 years ago, about a group of people whose plane crashes on an island and they can't leave the island? That's the idea—although language, unlike the show Lost, makes at least some sense, doesn't have a magical smoke monster...)

Here's an example of a sentence that allows normal movement, and a question we make from it:

  1. The silly story was important to Greg.
  2. The silly story was important to who?
  3. Who was the silly story important to _____?

That sentence is just like what we've seen before: we can turn a key word (like Greg) into a wh- question word, and move it to the front of the sentence.

Now let's look at a sentence which has an island, which a wh- word cannot escape from:

  1. The [silly story about Greg] meant a lot.
  2. The [silly story about who] meant a lot?
  3. *Who did [the silly story about ____] mean a lot?

If you ask any native English speaker, they will probably say that sentence #3 ("Who did the silly story about mean a lot?") does not sound like good English. It sounds weird, wrong, ungrammatical. There's no problem with the meaning, though; as we can see from sentence #2, which is acceptable, this is a perfectly reasonable kind of question to ask. But the problem is, "silly story about Greg" is a noun phrase island: for weird reasons that aren't very well understood, wh- words can't leave this island. Therefore, asking "The silly story about WHO meant a lot?" is fine, but if you try to move "who" out of the island and to the front of the sentence it becomes terrible. That's an island.

Here's another example of an island that wh-words can't escape from. While the previous example was a noun phrase island, this one is a relative clause island

  1. You read the book [that Greg wrote].
  2. You read the book [that who wrote]?
  3. *Who did you read the book [that ______ wrote]?

To me, this example sounds even stronger than the noun phrase island example. In this example, sentence #2 ("You read the book that WHO wrote?") sounds totally fine, and sentence #3 ("Who did you read the book that wrote?") sounds totally awful.

There's actually some debate over whether or not Chinese has islands, but there are some examples that look island-like to me. For example, we can say 我昨天 [跟曉明] 一起打羽毛球, a sentence where 跟曉明 forms a prepositional phrase; but we can't move 曉明 out of that prepositional phrase (when making a relative clause or something: "那個我昨天 [跟__] &#x;一起打羽毛球的很好", where "我昨天 [跟曉明] 一起打羽毛球" has been turned into a relative clause and 曉明 is turned into a gap, sounds awful. To take another example: some people consider a sentence like 那份李四說因為累還沒寫__的作業就在那兒 (where 作業, the object of 寫, has moved out of the relative clause) to be acceptable, but consider 那份李四说他 [如果寫了__] 就看報紙的作業就在那兒 to be quite bad because it seems like 作業 cannot move out of the 'if' clause. However, not everyone agrees with the judgments about these sentences, and not everyone agrees that the problems with these sentences are because of island effects or because of something else. Anyway that's not important for our present purposes, but you might find it interesting.

Parsing and islands

What does all this weird island stuff have to do with psycholinguistics?

Remember how the filled gap effect works: you are reading a sentence, your parser expects there to be a gap somewhere, and then you get surprised and confused if a word appears in the place where you expected that there would be a gap.

But what would happen if there's an island? If your parser is really good at grammar, it should know that there cannot be a gap in an island: if you saw a filler such as what, you know that cannot have come out of an island (because nothing can leave islands). Imagine you find a lost kid on the street in Central, and you want to help the kid find their way back home. You probably wouldn't check for their home on Cheung Chau, because a little kid would not know how to swim from Cheung Chau to Central; any kid you find in Central is probably not lost from Cheung Chau.

So, if your parser is really smart and pays attention to complex grammatical rules at the same time it's trying to understand sentences, it won't look for gaps within an island. When you read a word like what you will try to figure out where it came from, but you will "pause" that process if you start reading something that is a noun phrase island, a relative clause island, or some other kind of island.

The other possibility is that your parser is not so grammatically accurate, and just looks for gaps anywhere it can (and if it makes a mistake of predicting a gap inside an island, it might get confused and have to re-analyze the sentence structure later).

That is the question addressed in this paper: Stowe1986.pdf 

Stowe predicts that there will be a filled-gap effect in sentences that don't have an island. To see if the parser is grammatically accurate, she also examines sentences with islands. If the parser is grammatically accurate, the filled gap effect should disappear when the potential gap is in an island (because the parser is smart enough to know not to expect any island there). On the other hand, if the parser is grammatically dumb, there will be filled gap effects inside of islands, just like outside of islands.

Browse this paper (focusing on Experiment 2; I have added notes in the PDF to help guide you) to figure out what Stowe's experiment found. Make sure you are able to describe the kinds of sentences she used to test this question, what results she found, and whether the results support the theory of a grammatically accurate parser or a grammatically dumb parser. As always, you may read and discuss the paper with classmates.

Do the findings from Stowe (1986) support the idea that the parser is grammatically accurate? Why or why not?

Write a reflection on what you learned in this module.

When you finish this activity, you are done with the module (assuming all your work on this and the previous tasks has been satisfactory). If you are interested in leading a discussion on this module, you can go on to see the suggested discussion topics. Otherwise, you can return to the module homepage to review this module, or return to the class homepage to select a different module or assignment to do now.


by Stephen Politzer-Ahles. Last modified on 2021-07-14. CC-BY-4.0.