Assignment 3 grading and comments
On this page
Problem 1
Grads 55, Undergrads 40, a - 10 marks per sentence, total 40,
b(grads only) - 3 marks per item, total 15
A common problem for all trees was incorrectly specifying feature
values. Note that PP does not have case, agreement or other features,
except for pform and wh.
Another difficult question is recognizing PP's and
adverbials. Phrases such as "where" can correspond to either PP's or
true adverb phrases, such as "there". It it therefore possible to
analyze them as PP's, though they are adverbials from the point of
view of semantics, and phrases like "at home" can coordinate with
adverbs like "there". Therefore it is better to analyze such PP's as
adverbial modifiers, using the rule
ADVL -> PP[pred]
I marked the cases when the PP analysis was adopted, but didn't take
points off for that. The only truly incorrect analysis that someone
adopted was
PP -> ADVL
a
Almost everyone did the question correctly. Most common mistakes were
incorrectly specifying feature values for nodes.
b
People did generally well on that, though not everyone identified "The
key you mislaid" as the subject NP containing relative clause.
c
The most important part of the question was recognizing symmetry that
exists in the sentence: both verbs take an adverbial (or PP if you
wish) complement, so both should have the gap of the same type. Not
everyone noticed that.
d
Most people were thoroughly confused by this sentence. Note that the
rule for comparative structures states
NP -> NP[+er] than XP[/NP]
where XP can be S, PP or other NP. So the phrase sentence should be
analyzed as
[S Phil bought [NP [NP more jewelry] than [S [S Fran wanted t] or [S he could afford t]]]]
NP-> NP than S/NP
Problem 2
33 marks, 4 per sentence, 5 for conclusions
One important thing to note there is that what results you get for RA
and MA is the matter of syntactic analysis you adopt. There is no
exact answer, and different theories produce different results. There
are competing analyses of the sentence depending on how seriously you
take Xbar theory and which rules you adopt for prepositional phrase
adjuncts (some theories allow them to modify the verb head, while
others don't), and how do you define verb complements versus adjuncts,
which is sometimes a difficult question. Therefore it is absolutely
necessary to draw phrase structure trees in your arguments to show the
exact structure of the sentence you're assuming. Some people got
results opposite from mine because they had different trees. I
accepted every analysis as long as it was plausible, but I took off
points if I could not figure out what exact phrase structure you were
assuming.
Problem 3
10 marks
Most people did very well of that. The most common problem was
assuming that "born" was a base form of the verb. It is apparently
not, but rather the past participle of "bear". The other difficult
question is whether "In which town" is the adverbial modifier that
modifies the whole VP (this was the analysis that I adopted) or it is
subcategorized for by the verb. You can make arguments for both cases,
so I considered either analysis correct.
Problem 4
Undergrads - 15, grads a-7, b-3, c-6, d-4, total 20
People generally did very well on that problem, too.
Problem 5
a - 6, b - 2 marks per item, total 12
Everyone did very well on that, except maybe for item 2 in b. Note
that A and B being dependent(independent) does not mean that they will
be dependent(independent) given C.
Consider the following situation: A and B are independent events that
can cause C, for example, both spilled oil and soap on the bathroom
floor can cause you to slip. But after you have already slipped, it is
conceivable to that knowing that there was probably soap on the floor
lowers the probability that oil was spilled, too, so P(A|B,C) <> P(A,C).
Similarly, A and B an be dependent if A causes C and C causes B,
like when slipping can cause you fall down, and falling down can cause
breaking a leg, so slipping and breaking the leg are not
independent. However, if we already know that you fell down, it is
possible to say that knowing that you slipped on the spilled oil
rather than tripped over something does not influence the probability
of your breaking the leg. Therefore A and B are conditionally
independent given C.
If you're interested, I can give you a good explanatory article
that deals with problems of events being conditionally independent
within the Bayes net framework.
Statistics
Here are the scores so that you can compare your performance to
others.
(grads max 130, undergrads max 110)
51.50
92.00
97.00
98.00
122.00