Final

Due Tuesday 12/13, to CollectIt by 5pm (pdf only).

This exam counts for 35% of your course grade.

No late finals will be accepted.

Note: You may not discuss the final with other students (until after everyone has turned it in). Please send any questions about the final directly to me (via email) rather than posting on GoPost. I will respond by email and post answers as appropriate on the GoPost. Please read GoPost.

Problem 1. A Tree (60pts)

A. Draw a tree for the following sentence, using the grammar from the textbook and according to the instructions below. (NB: This problem is not only about long distance dependencies, but also about multiple phenomena involving the SPR feature and the rules and principles of our grammar.)

• This device everyone believes to be expensive to fix but impossible to replace.

Instructions

1. Show the whole tree (i.e., no triangles under nodes).
2. Assume lexical entries for the words instantiating appropriate types from the grammar. Specific tips:
   • There is one orv-lxm.
   • There are two easy-type adjectives.
   • Four different phrase structure rules are used, one seven differnt times.
   • You are not asked to show ARG-ST values, but it may be helpful to work out what they are.
3. Label each node with an appropriate abbreviation (e.g., NP).
4. Show the values of SPR, COMPS, GAP, and STOP-GAP where they are non-empty. These values should include tags where they are identified with other structures in the tree (whole nodes or the value of SPR, COMPS, GAP and/or STOP-GAP on other nodes). At least one instance of each tag should be associated with a category abbreivation (NP). If the grammar requires coindexing (but not complete identification), show that with a subcript (NP_i).
5. Please do NOT show the value for any other features. The tree is big enough as it is!
6. Check your work: The correct tree has 24 nodes, not counting the leaves (word forms). I found 35 non-empty values for SPR, COMPS, GAP and STOP-GAP collectively. I used tags numbered 1-14, plus the index i. (You don't have to use the exact same numbers in the same places; this is meant to help you check that you have the right number of sets of identities.)

B. Describe each step in the cascade of identities that link the INST value of the device predication to the FIXED value of fix predication. Your answer should take the form of an enumerated list. Each item in the list should describe one step, i.e., state which features are required to have the same value and state which lexical entry, rule or principle of the grammar requires this.

I found 28 steps in this chain. You do not need to mention any lexical rules, nor any specific lexical types. Identities that come from either can be described as enforced by the lexical entries themselves. You should, however, account for identities required by the ARP (I found 9 of those). To start you off, here are the first and last, plus one from the middle

• 1. The lexical entry for device identifies its INDEX value with the INST value of its device predication.
• ...
• n. The Argument Realization Principle identifies the the first element of the ARG-ST of to with the sole element of its SPR list.
• ...
• 28. The lexical entry for fix identifies the INDEX of the second element of its ARG-ST with the FIXED value in its fix predication

Problem 2. Some Lexical Rules (40pts)

A. For each word in the following sentence, list the lexical rules involved in licensing its word structure. (Again, this question is cumulative, and not about any one particular phenomenon. Also, while you do not need to provide a tree for your answer, it might help you work through the problem to sketch one out.)

• Kim expected it to be understood that Sandy's problems can't be solved.

Check your work: Though there are 13 words in the sentence, they collectively represent 17 applications of lexical rules. One lexical rule is used 7 times.

B. Show the word structure for can't as it appears in this sentence. Include the effects of the lexical rule(s) it undergoes as well as constraints inherited from types (e.g., those on auxv-lxm and its supertypes, as well as those on word). You don't need to show the RESTR values of the elements of the ARG-ST and valence lists or any other information that is underspecified by the combination of the lexical entry and lexcial rules but rather `unifed in' from other parts of the tree.

C. How does our analysis capture the fact that be is required before understood in this example? That is, why aren't the following (with nothing there or the other semantically empty auxiliary) licensed?

1. *Kim expected it to understood that Sandy's problems can't be solved.
2. *Kim expected it to do understood that Sandy's problems can't be solved.