A case of English vs. classical logic

[honestrosewater]

If P, Q, and R are propositions,

1) ((P -> Q) & (P -> R)) <=> (P -> (Q & R)).

But let

P = I have exactly one dollar.
Q = I have enough money to buy a coffee.
R = I have enough money to buy a pickle.

and (1) fails (though in only one direction) under the English interpretation that, if, say, a coffee and pickle cost a dollar each, would make ((P -> Q) & (P -> R)) true but (P -> (Q & R)) false for the reason that I would need at least two dollars for (Q & R) to be true. What else is going on in this interpretation that makes (1) fail? I'm almost certain and is being used differently, but I think it might be some other things as well, like I referring to the same person and some assumptions about time, simultaneity, or such.

Just thought I'd share. I think it's interesting but don't have time to think about it just now. Maybe someone else has an explanation? Can anyone make everything relevant explicit?

 

[Skomatth]

One way to resolve the tension between the formal and informal reasoning is to say that you haven't properly translated Q and R.

These two sentences have equivalent meanings:
1)I have enough money to buy coffee.
2)Possibly, I could buy coffee.

Since they have the same meaning, they should both be translated the same way. The second sentence makes it clear that we are dealing with a modal proposition. I'll use * to meaning possibly. Thus you have

1) ((P -> *Q) & (P ->* R)) <=> (P -> (*Q &* R)).

Note that (*Q&*R) does not imply *(Q&R). I think this better captures the way we use P,Q and R.

 

[honestrosewater]

Ah, thanks, I hadn't thought of that. But I don't think that's the problem I'm thinking of. I think it's that this and has an added 'together' requirement. Compare

3) Coffee tastes yummy and pickles taste yummy.
3a) Separately, coffee and pickles taste yummy.
3b) Together, coffee and pickles taste yummy.

(3) implies (3a) but doesn't imply (3b). Suppose you let coffee c and pickles p be individuals and Yx mean x tastes yummy. In (3) and (3a), Y is a unary predicate, but (3b) seems to want to either make Y a binary predicate (Edit: eh, and I don't know what that would be or how it would make sense. Although together and separately as adverbs might be considered to modify the predicate, I'm not sure whether those classes and rules are valid or relevant. Hmm...) or name a new individual, say g, coffee and pickles (Ack, though if coffee and pickles was being treated as a single individual, taste should inflect for subject agreement, becoming tastes. So maybe it's somewhere in between. Oh, right, I think it's just a compound subject. Hah.). This latter and seems to function like an operation on your domain. (3a) seems like just an abbreviation of (3), having the same underlying structure. But (3b) seems to be an entirely different structure.

I think the problematic interpretation depends on at least a) the function of and and b) the agent being the same in P, Q, and R.
For (a), consider that ((P -> Q) v (P -> R)) <=> (P -> (Q v R)) and has no such problem. Indeed, ((P -> Q) & (P -> R)) => (P -> (Q v R)) removes the problem.
For (b) consider that changing anyone of the I's in P, Q, or R to a non-first-person noun removes the problem. Or, hm, maybe it doesn't. What do you think?

 

[Orefa]

It is less misleading if you repeat the full statements without combining the verbs. I have exactly one dollar so I have enough money to buy a coffee and I have enough money to buy a pickle. BUT: I may or may not have enough money to buy a coffee and a pickle.

Ambiguities in natural languages are why they are not used in formal logic and why they don't translate directly.

 

[Hurkyl]

Yay, I'm not the only one who thought that the implication still went both ways in the original question!

I just don't like natural language. That's all there is to it.

 

[honestrosewater]

Ambiguities in natural languages are why they are not used in formal logic and why they don't translate directly.

Yeppers, I am studying linguistics and logic. What do you think are the necessary differences between a "formal logic" and a formal model of a natural language?

 

[honestrosewater]

Yeah, I'm pretty sure treating it as a compound subject/object would work out. So one way of looking at it is that that instance of and is functioning as an opertation on the domain instead of as a connective. I wonder if looking at it as changing the predicate somehow would work too.

Does anyone else recognize the interpretation I'm takling about? I mean, do you guys naturally read it both ways?

 

[Orefa]

What do you think are the necessary differences between a "formal logic" and a formal model of a natural language?

Each word needs to have one and only one meaning, a single precise definition. This is not the case in any natural language. Natural languages are dynamic and words constantly change meaning (eg: gay). You cannot control this, and you don't want to control it either. If you had a single meaning per word then vocabulary would expand beyond reasonable.

 

[honestrosewater]

Each word needs to have one and only one meaning, a single precise definition.

You mean each word form needs to be associated with only one meaning? Why? If you can use other information to disambiguate the word forms, e.g., their functions in a sentence, you still can end up with one meaning per word; you would just have to broaden, or unnarrow, your concept of a word. The different meanings are themselves differences -- just represent those differences in the model. I don't really understand the reason for this requirement anyway. Formal logics don't need to be consistent or sound or anything else that I can think of that might tie in with this requirement.

Do you think there is some linguistic phenomenon that cannot be formally modeled?

This is not the case in any natural language.

You know every natural language?

Natural languages are dynamic and words constantly change meaning (eg: gay). You cannot control this, and you don't want to control it either. If you had a single meaning per word then vocabulary would expand beyond reasonable.

Sorry, I was thinking of a synchronic model, a snapshot of the language at a single moment in time. Still, some aspects of language do change regularly, and rules of language change have been discovered and used, so diachronic models aren't out either. Do you really think scientists can't model dynamic systems? If we were talking about physics instead of language would you have a different opinion?

 

[AKG]

The problem is that you're equivocating:

"I have enough money for a coffee and I have enough money for a pickle"

with:

"I have enough money for a coffee and a pickle"

 

[Orefa]

I probably misunderstand what you are asking. Based on your OP I assumed you were interested in the differences between the language of formal logic and natural languages like English and Chinese. But I read again the question to which I was replying to see my mistake: you actually asked for the difference between formal logic and a formal model (of a natural language). Of course, formal logic is a method and a formal model is a representation. But maybe you misspoke so I won't venture too far. I can still reply to a couple of things.

Regarding disambiguation you are correct, one word per meaning is not essential when a grammar retains context. A single word can have different meanings depending on context, but then the context itself must also be unambiguous. Resolution problems arise when more than a single context co-exist. If a context exists (eg. crime) and then a second context is added (eg. poverty) then words defined differently in both contexts need to be explicitly qualified. Practically, a simpler rule such as one meaning per symbol is more reliable than a more complex disambiguation method.

You asked if I know every natural language. Of course not, but a "natural" language evolves naturally, and natural evolution implies changes. With change comes ambiguity, at least between old and new meaning during the transition period.

 

[honestrosewater]

The problem is that you're equivocating:

[4] "I have enough money for a coffee and I have enough money for a pickle"

with:

[5] "I have enough money for a coffee and a pickle"

Okay, maybe I picked a bad title. I know that English and propositional or predicate logic are different languages. I was actually interested in analyzing the two different English propositions represented by (4). In the version of English that I speak, (4) is ambiguous and one meaning of (4) is equivalent to one meaning of (5). I was wondering what exactly about (4) allows the interpretation that is equivalent to (5). I figured out one solution already. Two underlying structures of

6) I have enough money to buy a coffee and I have enough money to buy a pickle.

are

6a) Pab & Pac

and

6bi) Pa(b*c)

or more clearly,

6bii) Pad

where d = b*c and & and * are the two different operations represented by the word form and; & is an operation on your set of formulas and * is an operation on your set of individuals. So I guess I was right about the and being different (same form, different function) and the agent, the individual a in the example, having to be the same. See what I mean now?

 

[honestrosewater]

Of course, formal logic is a method and a formal model is a representation. But maybe you misspoke so I won't venture too far.

"formal logic" was your choice. I wasn't sure what exactly you meant by it. To me, a formal logic is a kind of (formal) language. What do you mean by method?

Regarding disambiguation you are correct, one word per meaning is not essential when a grammar retains context.

I'm actually back one step. I wasn't asking about how one meaning per word can be acheived; I was questioning whether the one meaning per word property is even a necessary property of formal logics. I don't think it is or needs to be. I don't think anyone has intentionally designed a formal logic that doesn't have that property, but the only consequence that I can think of of not having that property is that it would probably introduce inconsistencies. But logics can already be inconsistent for other reasons, so big wup. I'm just looking for a reason to make that a necessary property of formal logics. Of course, I obviously don't know exactly what all you're counting as a formal logic -- I'm not sure what all I'd count as a formal logic either -- but I'm just curious about your ideas.

 

[Orefa]

To me, a formal logic is a kind of (formal) language. What do you mean by method?

The application of some rules. We may have similar ideas in mind. Logic can be seen as a language, a set of rules and symbols, or it can be seen as its application. This is similar to math: is it a language or the application of this language? When you apply logic to a problem then you follow a method dictated by the rules in question, by the grammar of the language you use. So I guess this is another example where natural language is imprecise since we seem to be talking about the same thing using different terms, or slightly different things using the same terms.

I was questioning whether the one meaning per word property is even a necessary property of formal logics. I don't think it is or needs to be.

Ah, well, this is an interesting idea. I never even questionned that anything formal must be unambiguous, it seemed self-evident to me (since I did not question it). Why bother to create something formal unless it is also clear... But I guess it is quite possible to have a formally ambiguous grammar where, for example, A & B means either A and B or A and not B. This would reduce to just A using "common" logic. Ambiguous grammars are not necessarily useful though, but I suppose they are not impossible.

What I personally talk about when I speak of formal logic is the application of a set of operations on boolean values. These are classified with mathematical precision. I don't speak of fuzzy logic without using the adjective.

 

[honestrosewater]

Ambiguous grammars are not necessarily useful though, but I suppose they are not impossible.

Not useful? Secret codes, anyone? Yeah, perhaps as self-contained systems, they aren't very useful. But when information outside of the language can be used to disambiguate that within, ambiguity can make the language more efficient (I think that's exactly why natural languages are ambiguous in places). For example, as you noted, ambiguity cuts down on the size of our vocabulary. You can also use it to effectively keep your communications secret (by keeping which interpretation you intended unknown).

 

[Orefa]

But a language is a means of communication. If it fails to permit communication then it is not a language but a noise generator.

The generation of secret messages can be independent of the language used. You could send just half a sentence now and the other half later in secret. In general, you send out part of a message but its meaning is not clear until you receive the rest, which may be the other half of the sentence, a key that unlocks the first part, or a context indicator as in your example. They still must be useable by the language in order to extract its meaning.

Of course another way to have secret messages is to keep the language itself secret or partly secret. Then messages can be exchanged in full sight of everyone who doesn't know the complete rules of the language. But this is a different thing.

 

[hypnagogue]

If P, Q, and R are propositions,

1) ((P -> Q) & (P -> R)) <=> (P -> (Q & R)).

But let

P = I have exactly one dollar.
Q = I have enough money to buy a coffee.
R = I have enough money to buy a pickle.

and (1) fails (though in only one direction) under the English interpretation that, if, say, a coffee and pickle cost a dollar each, would make ((P -> Q) & (P -> R)) true but (P -> (Q & R)) false for the reason that I would need at least two dollars for (Q & R) to be true. What else is going on in this interpretation that makes (1) fail? I'm almost certain and is being used differently, but I think it might be some other things as well, like I referring to the same person and some assumptions about time, simultaneity, or such.

Just thought I'd share. I think it's interesting but don't have time to think about it just now. Maybe someone else has an explanation? Can anyone make everything relevant explicit?

I think the tension you have here between the English interpretation and the formalism may be occurring simply because English has a deeper structure than propositional logic. If you take a strict interpretation of the sentences involved, there is no contradiction between the English interpretation and the formal interpretation above: P -> (Q&R) just reads "If I have a dollar, then I have enough money to buy a coffee, and I have enough money to buy a pickle," which is true so long as you don't interpret it to mean "I have enough money to buy BOTH a coffee AND a pickle."

In reading out that sentence in English, though, perhaps we do naturally want to read it as "I have enough money to buy both a coffee and a pickle," which in turn might imply that a first-order logic formalism is closer to how we naturally interpret the sentence. For instance, let

D(x): "I have x dollars"
B(y1,y2,...,yn): "I have enough money to buy y1 and y2 and ... and yn"

then strictly speaking we have, as above,

[(D(1) -> B(coffee)) & (D(1) -> B(pickle))] <=> [D(1) -> (B(coffee) & B(pickle))]

but perhaps on reading the English sentence, we naturally assume the "and" function is working inside B rather than outside, and thus interpret the latter clause as

D(1) -> B(coffee, pickle)

This causes a tension because, given the way we've defined our first-order sentences and given the way commerce works, it is not always the case that (B(x) & B(y)) -> B(x, y)

Gotta run for now, but hopefully that's some food for thought. I also have an inkling that the "xor" function could come in handy here, but I'll have to flesh out that intuition later.

 

[honestrosewater]

In reading out that sentence in English, though, perhaps we do naturally want to read it as "I have enough money to buy both a coffee and a pickle," which in turn might imply that a first-order logic formalism is closer to how we naturally interpret the sentence.

I came up with an answer similar to yours (see posts 3, 12). Only the arity of the predicate doesn't change; one and is operating on formulas, the other on individuals. It's just two different words with the same form, as the verb form and the noun form.

One theory: Verbs refer to events or situations that involve characteristic roles called thematic roles, or θ-roles. For example, a killing typically involves a party that performs the killing (the agent) and a party that is killed (the patient). The parties filling these θ-roles are called the arguments of the verb (and I think are typically represented in English by Determiner Phrases). The representation in the mental lexicon (the dictionary in your brain) of the meaning of a verb includes its number of arguments and its θ-roles. This is all semantic stuff, by the bye. In a sentence, syntactic rules assign the θ-roles their syntactic positions. This interaction between semantic and syntactic structures explains why

1a) Romeo killed Tybalt.

describes an event that is different than

1b) Tybalt killed Romeo.

but equivalent to

1c) Tybalt was killed by Romeo.

-- in (1a) and (1b), the agent role is assigned the subject position, while in (1c), the patient role is assigned the subject position. It also explains why

2) Romeo bought.

is ungrammatical -- bought needs another argument, the thing that Romeo bought, poison. Changing the arity of the predicate is the same thing as changing the number of arguments of the verb, which is changing the verb's θ-roles (part of the meaning of the verb). This is what from does in

2a) Romeo bought poison from an apothecary.

where bought has 3 arguments, (Romeo, poison, an apothecary), but and fails to do in

2c) Romeo bought posion and an apothecary.

where bought has 2 arguments, (Romeo, poision and an apothecary).

I think the second interpretation of the original sentence, with and operating on individuals, might result from a speaker's zeal to remove (perceived) redundancy. It's actually the first way that I read it. Well, it was first by a fraction of a second, but still.

 

[Skomatth]

Honestrosewater,

Can you recommend an intro book on semantics?

 

[honestrosewater]

Can you recommend an intro book on semantics?

Sorry, I haven't studied much semantics yet. I might still be able to help a bit though. Do you (know whether you) want a formalist or a functionalist approach? Have you already been introduced to semantics in a survey course (e.g., you're familiar with compositionality, entailment, presuppositions, assertions, how truth conditions relate to meaning, what a theory of meaning is supposed to be, and such)? How much logic do you know (up to predicate/first-order)? From what I've seen, semantics is basically just logic, but my book's coverage of semantics is paltry.

Books that I see recommended and used most often:
https://www.amazon.com/gp/product/0631197133/?tag=pfamazon01-20
https://www.amazon.com/gp/product/026253164X/?tag=pfamazon01-20

For logic for linguists, these are good so far and came highly recommended:
https://www.amazon.com/gp/product/0226280853/?tag=pfamazon01-20
https://www.amazon.com/gp/product/0226280888/?tag=pfamazon01-20

Personally, I think this one looks good:
https://www.amazon.com/gp/product/0198700105/?tag=pfamazon01-20

You might want to search the descriptions and reviews in LinguistList's http://cf.linguistlist.org/cfdocs/new-website/LL-WorkingDirs/pubs/index.html if you haven't already.

 

[matt grime]

It also explains why

2) Romeo bought.

is ungrammatical -- bought needs another argument, the thing that Romeo bought, poison.

I don't think you'll find that is correct.

Consider the following fragment of conversation:


R: We were out for dinner. I bought.


It is perfectly valid. And allowiing for transitive and intransitive forms of verbs is danagerous. The following is prefectly acceptable in the US to mean that Romeo wrote a letter to Juliet yesterday but has a completely different meaning to any other native English speaker.

R: Juliet wrote me yesterday.

 

[mrj]

Hi. You can also check out this book: "Everything that Linguists have Always Wanted to Know about Logic...But were Ashamed to Ask," by James D. McCawley, available on Amazon.

mrj

 

[honestrosewater]

Woot!

R: We were out for dinner. I bought.It is perfectly valid.

Sure, sorry, any given sentence (or other unit) might be acceptable to some speakers and unacceptable to others, but for IMO obvious reasons, we don't want to let a sentence be both grammatical and not grammatical at the same time. It seems to me that a sentence is grammatical iff it's in the set of sentences that you're trying to model at the moment (or perhaps also iff it is probably grammatical based on the data from the speaker(s) whose mental grammar(s) you're trying to describe). That's the way I use it anyway. Of course, that set is usually infinite, so perhaps you just have to decide whether or not a certain sentence or sentence form is grammatical as you go along. I don't really know. Grammaticality is relative though. Anywho...

I think

R) We were out for dinner. I bought.

is grammatical and bought does have two arguments in (R). The second argument is dinner, but it is dropped, or it has a zero form, just as the second know has a zero form in

They don't know, but I do (know).

If I told people (R) and asked them where we went to dinner, I would expect most of them to answer, in so many words, that they didn't know or the sentences didn't say, and I think the sentences really don't say. But if I asked them what I bought, I would expect most of them to answer, in so many words, that I bought dinner. I expect this because I think (R) does imply

S) I bought dinner.

(i.e., (S) is part of what (R) 'means'). I think the second dinner is dropped because when people aren't in a rush, they're lazy, people communicate cooperatively, bought does require two arguments, and when you don't bother to supply the one filling patient role, the listener cooperatively goes looking for something eariler in the conversation that could fill it (similar to looking for the referent of a pronoun). That kind of thing happens a lot. I just include a rule in my grammar that allows the unit in the second dinner's position to have a zero form under conditions satisfied by (R), and on the semantic side of things, I have that (R) is equivalent to

T) We were out for dinner. I bought dinner.

I'm certainly open to other analyses though. I personally don't know an English word of the form bought that only needs one argument and don't recall ever seeing it used that way.

And allowiing for transitive and intransitive forms of verbs is danagerous.

Why is that?

The following is prefectly acceptable in the US to mean that Romeo wrote a letter to Juliet yesterday but has a completely different meaning to any other native English speaker.

R: Juliet wrote me yesterday.

One theory might explain this with selectional restrictions and the possible syntactic assignments of θ-roles. According to this theory, the lexical entry of a verb can include what are called selectional restrictions, which restrict the types of things that can fill each θ-role (e.g., by naming properties that it must have). For example, someone's entry for the verb drink might include the restrictions that the agent be animate and the patient be liquid. By the bye, this would be for the 'normal' meaning; words can be used rhetorically, of course, and presumably, such use is based on the normal meaning.

Write, as I use it, can have at least three θ-roles: 1) an agent, the thing doing the writing, 2) a patient, the thing written, which must be inanimate, and 3) a goal, the destination if the action involves motion, or the recipient if the action involves transfer of possession. The problem seems to be that

1) Romeo wrote me yesterday.

can be interpreted with Romeo as the agent and
-2) yesterday as an adverb and
--a) me as the patient or
--b) me as the goal.
-3) me as the goal and yesterday as the patient.

To me, (2a) violates a selectional restriction because my rules say that me is animate and the patient of write must be inanimate.* (3) can be wrong because a) yesterday, as the common noun meaning the day before today, also isn't something that can be written (so it may also be considered to violate a selectional restriction, though I'm not sure how best to describe that) and is wrong also because yesterday is the singular form of a common, count noun and that slot requires that if it's filled with a common, count noun, that noun must be plural, or b) as written, yesterday should have been capitalized if it was a proper noun. In speech, (3), with yesterday understood to be a proper noun, is acceptable but (2b) is preferred over it. I imagine I would read (1) as (2b) unless something wrtten and named Yesterday was already being talked about.

It might be that other speakers reject (2b) because they require, for that construction, that if the goal role appears in the sentence, the patient role appears in the sentence. That is, with the agent role assigned to the subject position, some possibilities are

4) [agent] wrote. (Romeo wrote.)
5) [agent] wrote [patient]. (Romeo wrote a letter.)
6) [agent] wrote [goal]. (Romeo wrote Juliet.)
7) [agent] wrote [patient] to [goal]. (Romeo wrote a letter to Juliet.)
8) [agent] wrote [goal] [patient]. (Romeo wrote Juliet a letter.)

and they don't accept (6). Is that what's wrong with it to you? I'm curious about this now.

*Sorry, rather, the referent of me is animate, but eh, I'm lazy... and I probably made other similar kinds of slips elsewhere too.

 

[honestrosewater]

Now that I think about it, I suppose I might have had, in some sense, a bought with only one argument since I had buyer, as in

9) Bob is a buyer.

which implies (and could be used to construct)

10) Bob has bought.

(Edit: and (10) follows by the rules (I think) I already have that [Determiner Phrase] is a [verb]er implies [DP] [verb]s and ... wait, (10) is not Bob bought. The auxiliary has might make a difference... maybe there's a habitual/continuous action thing going on... hmmm... anywho...)

But one might analyze (10) as

11) Bob has bought (something).

i.e., as having a zero form for the second argument. (9) also implies

12) Something has been bought by Bob.

which is equivalent to and a common way to paraphrase (11). But I'm not sure how keen I am on the whole idea of zero forms. At any rate, I'm considering changing my entry for bought.

Edit: I think some actions or situations might just typically require at least a certain number of participants/arguments, but perhaps when an action is treated more generally, e.g., as continuous or recurring, the number of arguments can change... because it's not really the same action but some kind of collection of that action. Dunno, just thinking...

 

[honestrosewater]

matt, if you're still there, (or anyone else) just out of curiosity, what do you think of

Romeo wrote to Juliet.

?? Is it acceptable? Is it the same, better, or worse than

Romeo wrote Juliet.

??

 

[selfAdjoint]

Both are perfectly acceptible. I think the second one would sound better with a direct object: Romeo wrote Juliet a letter, or Romeo wrote Juliet that he loved her.

 

[loseyourname]

matt, if you're still there, (or anyone else) just out of curiosity, what do you think of

Romeo wrote to Juliet.

?? Is it acceptable? Is it the same, better, or worse than

Romeo wrote Juliet.

??

If the two are meant to be equivalent statements, then the first is a better way of stating it, since in this case, "Juliet" is dative and not accusative. Putting the "to" in front of her name makes that explicit and removes the ambiguity.

 

[honestrosewater]

Both are perfectly acceptible. I think the second one would sound better with a direct object: Romeo wrote Juliet a letter,

Yeah, two constructions (in English) for a ditransitive verb (a Verb having a Subject, Direct Object, and Indirect Object) are the

Prepositional Dative Construction (PDC): [V] [DO] to [IO]

(so called, I figure, because the dative case marker is associated with the IO position (esp. cross-linguistically), which occurs as the object of the preposition to (so you might consider to to be a dative case marker in a sense, or maybe not, I should stop thinking out loud so much hah) (and I think the IO position being filled usually implies S and DO are already filled).) and the

Double Object Construction (DOC): [V] [IO] [DO]

So you could look at

13) Romeo wrote Juliet.

as

a) a PDC with the DO and to dropped,
b) a DOC (what you seem to want to turn it into) with the DO dropped, or
c) wrote as an intransitive verb (only has a S) with two arguments: i) an agent participant, Romeo's referent, which occurs in the DP Romeo in the S position, and ii) a goal participant, Juliet's referent, which occurs as the object in the Prepositional Phrase (PP) to Juliet with to dropped (having a zero form).

I think (c) works. Um, and in case anyone is interested, it's kinda complicated, but consider some examples with a verb that "can't" be used intransitively, like buy, sell, send, and remember that you want Juliet to be the goal, not the patieht.

14a) *Romeo bought Juliet.
14b) *Romeo bought for Juliet.
14c) *Romeo sent Juliet.
14d) *Romeo sent to Juliet.

So why does wrote work? Because it can be -- and is being -- used intransitively, as in

15) Romeo wrote.

Compare to

16) Romeo sang.

and add the PP to both

17) Romeo wrote to Juliet.
18) Romeo sang to Juliet.

But for some reason

13) Romeo wrote Juliet.
19) *Romeo sang Juliet.

I haven't figured out yet under what conditions to can have a zero form -- it might just be an idiosyncracy of wrote. If anyone wants to examine some examples, have at it. You're looking for a verb that is (in the particular case)
a) intransitive with
b) two arguments,
i) an agent, occurring in a DP immediately before Verb
ii) a goal, occurring in a DP in a PP immediately following Verb
So the sentence is of the form [DP Verb PP] and you want to compare when [DP Verb Prep DP] and [DP Verb DP] are grammatical. I really want to know if there's a rule. (A linguistics LaTex package would rock too.)

or Romeo wrote Juliet that he loved her.

I'm not sure that there is a DO or IO position in Romeo wrote Juliet that he loved her. I think that is functioning as a Complementizer here, embedding he loved her, and I'm not sure that a Complementizer Phrase (CP) can occupy an object position (I think perhaps only DPs can occupy object positions). And if Juliet is here actually the PP to Juliet with the to dropped, then wrote is here intransitive. The referent of Juliet is the goal participant, but Juliet is not the IO. I'm still very much learning though, so who knows. I don't see why it wouldn't make sense to assign a CP a single θ-role, which I guess would be the patient role here, but I don't know. Bleh.