Linguistics 567: Grammar Engineering

Lab 1: Getting familiar with the LKB

Due 4/1/05

Getting Started

Download the .emacs file, and save it in your (ling567's) home directory.
Create a directory for yourself within ~ling567:
```
cd
mkdir /username/
```
Download the starting grammar for this exercise, and save it in your directory. The expand the .tgz file:
```
tar xzf grammar1.tgz
```
You should have a directory now called "grammar1".
Run emacs:
```
emacs &
```

M-x allegro
M-x fi:common-lisp
[hit return until it stops asking]

Load the gramamr: In the LKB Top Menu, select Load > Complete Grammar. Navigate to your-directory/grammar1 and select script.
Parse the sentence "the dog chased the cat": select Parse > Parse input...
Click around a bit and explore. Be sure to try Parse > Show parse chart

Run a testsuite with the LKB batch parse utility

In the LKB Top Menu, select Parse > Batch Parse
In the file selection dialogue, select grammar1/test.items
In the next file selection dialogue, select grammar1/test.results
Say yes to overwriting it.
Switch to the *common-lisp* buffer in emacs to and wait for the "Finished test file message".
Examine test.results. The first number after each sentence is the number of parses. Ungrammatical (starred) sentences, should have 0 parses, grammatical ones 1 or more. The other number is the number of edges found.

Feature Structure Re-entrancies in Agreement

Open the file types.tdl and locate the definitions of pos and its subtypes.

Modify that hierarchy so that the feature AGR is not appropriate for prepositions.

Now open the file lexicon.tdl and locate the lexical entries for the nouns. Note that they stipulate their own AGR and their specifier's AGR separately.

Modify these entries so that they require identity between their own AGR values and that of their determiners. For reference, the identity "tags" in tdl look like this:
```
x := y &
[ F #1 & z,
  G #1 ].
```
This definition means that the value of the feature F is z and the value of the feature G is the same as that of F. (For more on tdl syntax, see
this FAQ.)
Verify your changes with the testsuite test.items.
In your write-up, describe what you have done for this part. What changes did you have to make?

Phrase Structure Rules

Your answers to the questions below should be included in your write-up.

Find a sentence that parses with the verb give.
How does the parse assigned by this grammar differ from the one you would expect from the grammar in SWB 2003?
Examine the head-complement rule in rules.tdl. What about this rule is behind the difference noted above? How does it ensure that heads only get the complements they're looking for, and that the order on the COMPS list is obeyed?

Modification

This grammar does not yet have any account of modification. In this part, you'll add PP modifiers and the types and rules that support them.

Add a feature MOD to pos. Constrain each subtype of pos to have an appropriate value for MOD. Notes:
- The value of MOD should be a list.
- When the list is non-empty, it should contain one item of type expression.
- PPs can modify both nouns and verbs. To capture this, consider modifying the hierarchy under pos to give noun and verb a common supertype.
Add a head-modifier rule to rules.tdl. Refer to the head-specifier rule that's there and SWB 2003.
Add an appropriate lexical entry for near to the lexicon.
Test your grammar with the testsuite in mod.items.
In your write-up, describe what you have done for this part. What changes did you have to make?

The Head Feature Principle

Looking at the various rules, you'll notice that the HEAD value of the whole phrase is always identified with the HEAD value of one of the ARGS, sometimes the first and sometimes the second. In this part, you'll modify the grammar to reflect this distinction between head-initial and head-final phrases.

In types.tdl, add three new types:

head-initial := phrase &
 [ HEAD #head,
   ARGS [ FIRST.HEAD #head ] ].

head-final := phrase &
 [ HEAD #head,
   ARGS < expression, [ HEAD #head ] > ].

root-head-final := root & head-final.

Modify rules.tdl so that each rule inherits from one of these types. Remove any now redundant constraints from those rules.
Test your changes with test.items and mod.items.
In your write-up, describe what you have done for this part. What changes did you have to make?
What happens if you remove the type root-head-final. Explore why this might be (hint: use grep in your linux terminal to see what other files mention root. In your write-up, describe what you think is going on.

Eliminating Redundancy in the Lexicon

The grammar so far lacks any lexical type hierarchy. We'll fix that now.

In types.tdl create types for noun-word, verb-word, det-word, and prep-word. These types should state what is is common across all the nouns, etc. in the lexicon.
Create subtypes of noun-word for singular and plural nouns (note that this grammar is treating these as separate lexical entries, not related to each other by lexical rule). Update the lexicon so that the nouns inherit from these types.
Create subtypes of verb-word for (present tense) verbs that take 3sg and non-3sg subjects. Update the lexicon so that the verbs inherit from these types.
Update the lexicon so that the determiners and prepositions inherit from det-word and prep-word.
Test your grammar against test.items, mod.items, and agr.items.
Create subtypes of verb-word for intransitive, transitive, ditransitive, and prepositional transitive verbs.
Take advantage of multiple inheritance to cross-classify the valence pattern subtypes with the agreement subtypes. Hints:
- tdl notation for multiple inheritance:
```
x := y & z & [A b].
```
- Lexical entries (instances) should only belong to one type. Create the `marriage' types in types.tdl (3sg-intransitive, etc).
Update the lexicon file so that the verbs inherit from the new types.
Test your grammar against test.items, mod.items, and agr.items.
In your write-up, describe what you have done for this part. What changes did you have to make?

Save your work in progress

cd ~/username
tar czf grammar1.tgz grammar1
scp grammar1.tgz username@dante.u.washington.edu:~/

To download your work in progress and start up again:

cd ~/username
scp username@dante.u.washington.edu:~/grammar1.tgz
tar xzf grammar1.tgz

Submit to E-Submit

This lab is notionally due Friday evening. Catalyst will close the assignment at midnight Sunday night.

Create one compressed file (preferably with tar, as above), including both your grammar and your write up. Write ups can be plain text or pdf.

The link for our E-Submit site can be found on the course web page.

Return to main course page