[moschles]

Are there any other wild alternatives to semantics, other than conversion to First-order Predicate Logic? To understand what I am asking, read the description below.

SOME BACKGROUND

• Beginning around 1983, Ai research began to handle the problem of semantics in natural language. This project eventually came to be known as CYC. This research kicked off an established paradigm of representing the meaning of english sentences as a corresponding formula written in first-order Predicate logic ('FOPL').
• Even today, with modern PLNs, and weighted Markov networks, the foundational paradigm is still converting all english sentences into a corresponding formula in FOPL. 
• Consider chapter 12 within Artificial Intelligence: A Modern Approach 2nd ed. (2010) Here the author (Norvig) simply presumes that meaning of any english sentence should be appropriately captured by converting it into FOPL.
• Norvig repeats this paradigm again in chapter 22 and 23 of the same text.
• In section 12.5.1 Norvig makes the audacious claim that semantic networks are themselves a form of logical network, further reinforcing the semantics-as-FOPL paradigm.

THE ALTERNATIVES

• Consider the wild idea that the rules governing semantics in natural language is not the same as the rules that govern predicate logic.
• Semantics-as-FOPL makes a presumption at its core. Namely, it presumes that the words appearing in a sentence are instantiations of a particular member of a category. This may be completely false in the way humans understand language. Yes, I am claiming that particular words in a sentence are not naming the member of a category.
• On Imagining alternatives. We want to imagine that english utterances create categories spontaneously, that did not previously exist prior to the speaking, and reception of the sentence. The meaning of a sentence in english becomes the place where prototype categories overlap, likened unto venn diagrams. The number of dimensions in the category diagram is equal to the number of parts-of-speech which appear in the sentence.
• Adult language is very sophisticated, and it contains subordinate phrases. The function of subordinate phrases is the creation of categories in mid-sentence.
• The semantics of the sentence The boy kicked the ball. is actually a singular object with its own categorical properties. Namely the semantics is a boy kicking a ball. The verb is tensed to the past only because that is the method of locating the sentence in a temporal narrative. That is, natural language has used the social method of tensing verbs to locate it in a narrative, but the categorical aspects are orthogonal to this 'verbal' tool. 
• Imagine that semantics is not the denotation of truth value, but is instead the process of spontaneous category creation in realtime. That sounds spooky and crackpotterish at first. But to see why this is probably true, perform this exercise by yourself with a magazine. Find a noun from one article in the magazine. Find a verb from another article, and find a second noun in a third article. Combine these three words, so that you get a sentence of the form X verbs Y. You will find, as I did, suprisingly: that in 95% or more of the cases, the sentence that you form will make perfect semantic sense. That is, it is extremely difficult to formulate a random sentence that has perfect grammar and absolutely no meaning. (the infamous "Green ideas sleep furiously.")
• From a magazine on finance and economics, I formed the sentence The thunder called upon the ninja. This was totally unexpected from a magazine of this type, but the strangest thing is that the sentence makes perfect semantic sense.
• The existence of (rare) sentences which are grammatically perfect, but whom do not make any sense is a strong indication that the way human brains understand semantics is not by the rules of FOPL. There is some other process going on , of which the academic Ai community is not yet aware.

 
So we come full circle and back to my original question:
Are there any other wild alternatives to semantics, other than conversion to First-order Predicate Logic?

[Don Patrick]

I don't have enough formal knowledge about NLP methods, but I've been tinkering about on my own and found that
A. Grammar is just a guideline of preference that we formally agreed on, its rules approximate but differ from the rules through which we understand language.
B. Dependency on "good" grammar and complete sentences covers only half of all things written.
C. FOL is nevertheless a good starting place to get to grips with the basics of language, as long as you remember not to stick to it. Nouns are a particularly flawed category.

Lastly, there is a great deal of ambiguous and figurative speech in language, such as the thunder "calling upon" the ninja. To know that this is figurative, one has to know that thunder is not a tangible or living subject capable of performing that action literally. This is purely semantics/knowledge and can not be derived from language rules alone. But it does help to know that thunder is the subject and calling the verb.

[ivan.moony]

Maybe Type Theory, but you have to define operators and, or, not, etc., so basically it reduces to X order logic.