## Gregory Golberg Lab 1a 6.863 Spring 06

1. Question 1

The result is refering, where it should be referring. R is doubled as this is a case of gemination (Rule 15). In fact - the things you learn! - apparently this case is called West Germanic Gemination, in which:

```1) the suffix begins with a vowel (V)
2) base ends in a vowel-consonant (VC) pattern
3) the vowel of the above VC pattern is stressed before the suffix
Hence, refer - referring but not *referrence*
```

Thus, to create the correct generation, we have to indicate the stress. For the input "ref`er+ing", we get the following: 2. Question 2

"Dogs" is recognized as `dog+s [Noun(dog) + PL]. In order to recognize it also as a third-person singular present tense verb (from "to dog"), we add to the Lexicon and Alterations, under V_ROOT_NO_PREF:, `dog V_Root4 Verb(dog). Here's the result, with the change to Lexicon and correct output circled in red. 3. Question 3

We observe back-tracking, not a straight left-to-right pass. However, would like to split hairs and say that this statement is true, but not completely. If we take it to mean "a single non-deterministic pass", then it is true. We are simply observing an execution of non-deterministic algorithm on a deterministic machine.

Similarly, back-tracking also occurs during generation. However, since underlying forms are more "regular", there are less possibilities to try for recognition than there is for generation. Thus there may be more back-tracking during generation. In other words, we recognize both "boxes" and "oxen" as plural nouns, root + s, but "box + s" generates "boxes" (unless you're a geek) and "ox + s" generates "oxen". If this example were the entirety of our language, we need only one pass for recognition, but maximum of two for generation.

4. Question 4

As a result of rules applying rule12.rul in either order, the result is "kampat". There is no ordering effect (am I missing something?); the rules do apply simultaneously. It is precisely because they do, while rule 2 is "fed" (is that the right term) by rule 1, that rule 2 does not get executed. If the ordering did apply, then we'd get expected effect in the second case.

A solution, altering both automata, is shown below: As expected, here kaNpat -> kammat, and kampat -> kammat. As expected also, the reordering does not matter - below is the same result with the rules reordered: Using 2 automata is less linguistically transparent, but is more manageable. If the interacting rules were larger in number, each single automaton would grow more complex.