4

I'm reading Foundations of Statistical Natural Language Processing, and I'm doing one of the early exercises, trying to work out some of the language infliction about the word 'fun'.

  1. On the weekend the children had fun.

Trying to make a phrase structure parse of the above sentece, I'm not sure how to structure it. All formal grammars I've read describe a sentence as:

S → NP + VP

But I don't see how "on the weekend" could be a noun phrase? So far I've got this, but it doesn't seem right:

(PP On)(NP (D the)(N weekend))(S (NP (D the)(N children))(VP (V had)(NP (N fun))))

Resulting in this parse tree:

parse tree

So my question is: What's the correct Part-of-speech tagging for (1)?

2
  • I don’t understand what you think is amiss with the current POS assignments. Why would “the weekend” be anything but an NP? Is the problem that you don’t know to which element the whole PP attaches/applies/modifies? It’s a “when” phrase, which usually attaches to a VP.
    – tchrist
    Commented Dec 11, 2012 at 12:26
  • Well the weekend is obviously an NP, but I was missing the fact that the on/IN would give me a prepositional phrase PP. Commented Dec 11, 2012 at 15:21

1 Answer 1

4

EDIT: Given the sentence:

On the weekend the children had fun.

You can get a dependency parse (described at the bottom of this posting) that looks like this:

dependency graph

Which I believe may be more of what you are looking for.


The Berkeley parser produces this parse using its simple online interface:

(ROOT
  (S
    (PP (IN On)
      (NP (DT the) (NN weekend)))
    (NP (DT the) (NNS children))
    (VP (VBD had)
      (NP (NN fun)))
    (. .)))

Which can be diagrammed this way:

sentence 1 parse tree

On the other hand, if you rearrange the sentence slightly:

The children had fun on the weekend.

You get this tree:

(ROOT
  (S
    (NP (DT The) (NNS children))
    (VP (VBD had)
      (NP
        (NP (NN fun))
        (PP (IN on)
          (NP (DT the) (NN weekend)))))
    (. .)))

Whose diagram is this:

parse tree for sentence 2

For a simple parse, I think that is as good as you are going to get it. But you wish to try a constituency parser that can show linkages more complex that the simple tree above illustrates. For example, using CMU’s Link Grammar Parser:

    +------------------------Xp-----------------------+
    +---------------Wd--------------+                 |
    |      +-----------CO-----------+                 |
    |      +----Js---+              |                 |
    |      |  +--Ds--+      +--Dmc--+---Sp--+--Os-+   |
    |      |  |      |      |       |       |     |   |
LEFT-WALL on the weekend.n the children.n had.v fun.n . 

Notice the CO link applies that opener to the entire rest of the sentence. If you follow the [CO link]’s docs(http://www.link.cs.cmu.edu/link/dict/section-CO.html), you find that this is “sentence opener” link, used to connect sentence openers to the subjects of sentences. It mentions, amongst other things:

Openers may take commas; almost all words with CO+ therefore have "({{Xd-} & Xc+} & CO+)". With participles and adjectives, the comma is obligatory; "*Still upset about Joe they went to a movie" seems wrong. The Xd- allows a comma before the phrase as well as after. This frequently happens if the opener is not at the beginning of the phrase: "They claimed that, on Tuesday, they went to a movie." If the opener begins the sentence, then a comma before the opener is of course incorrect, but we allow it. See "X: Comma phrases".

And indeed, if you add the comma, you get another element in the parse (the Xc element), but nothing else changes, showing that these are equivalent:

    +-------------------------Xp------------------------+
    +----------------Wd---------------+                 |
    |      +------------CO------------+                 |
    |      +-------Xc------+          |                 |
    |      +----Js---+     |          |                 |
    |      |  +--Ds--+     |  +--Dmc--+---Sp--+--Os-+   |
    |      |  |      |     |  |       |       |     |   |
LEFT-WALL on the weekend.n , the children.n had.v fun.n .

For more details, check out the Standford NLP Group’s page here. I have used some of those tools, and they take a fair bit of set-up that I would never wish on a non-programmer, but they can be quite interesting.

You may wish to also try a dependency parse. These can be harder to read, but provide better linkages. One dependency parse visualization tool can be downloaded from here.

If you jump through all their hoops, you get the following output:

dependency graph

Notice that at last we can see that the PP at the start of the sentence correctly applies to the VP.

5
  • Thanks for the answer, it hadn't even occurred to me that I could just rearrange the sentence to get a much better understanding of it. I was looking for a phrase structure parse, and not a dependency parse, but you might be right that it's more expressive. Commented Dec 11, 2012 at 15:20
  • BTW: Why is Berkeleys parser making those (...) markings? Commented Dec 11, 2012 at 15:22
  • @Saebekassebil If you mean the (. .) at the end, it is for the final stop / period at the end. If you just mean parens in general, it’s because that’s the way people write their trees in NLP.
    – tchrist
    Commented Dec 11, 2012 at 15:25
  • Haven't got up to transformations, yet, eh? See here for an abbreviated list. Commented Dec 11, 2012 at 16:26
  • 1
    The DependenSee link gives a 500 at the moment but the source for it is here.
    – Jason C
    Commented Dec 20, 2015 at 3:29

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .