Ambiguity as a Species Defect
Ambiguity in natural languages is commonly regarded as a lapse from perfection. A perfect language would not contain ambiguity. Why is this? Because language is used for communication and ambiguity impedes communication. If an utterance is ambiguous, it is harder for the hearer to figure out the intended meaning; and in many cases it is not possible to do this without further questioning. Language is then failing in its purpose (or one of them), which is to convey information quickly and effectively. Ambiguity is the enemy of understanding. If we had invented language from scratch, or were constantly reinventing it, we would be thought guilty of poor craftsmanship—creating a defective product. Ambiguity is clearly not a necessary and unavoidable feature of language, since invented languages are often designed to be free of it. We can construct languages that contain no lexical ambiguity or syntactic ambiguity, as with standard formalized languages. So the defect of ambiguity is a contingent feature of natural human languages not a necessary feature of languages as such. 
Nor is the problem local or confined; a typical human language such as English is rife with ambiguity. Often we don’t notice it because the intended reading is so salient, but the formal structure of the language generates ambiguity all the time. The classic “I shot an elephant in my pajamas” is ambiguous in a characteristic way, i.e. it is not clear whether the modifier “in my pajamas” applies to the speaker or the elephant. The sentence “Old friends and acquaintances remembered Pat’s last visit to California” is said to have 32 different readings.  The Chomsky favorite “Flying planes can be dangerous” has infinitely many counterparts (e.g. “Dating women can be dangerous”). Syntactic ambiguity is pervasive and prodigal. Thus we must be constantly on our guard against it for fear of failing to express our meaning. Language is an ambiguity trap that easily lures us into error. We are always in danger of failing to communicate given the formal nature of the vehicle. It could have been worse—our every utterance could have been dogged by ambiguity—but things are bad enough as it is. And yet ambiguity is not integral to the very nature of language. Apparently we have been sold a shoddy product, one expressly constructed to get in the way of communicating. 
Why is this? Why is human language so defective? The question acquires bite when we acknowledge that language is a biological phenomenon: it is an adaptation shaped by natural selection, encoded in the genes, part of our birthright as a species. It is as if we have all been born with a defective heart or liver that does its job only fitfully and inefficiently. True, our bodily organs are not perfect—they can become diseased and break down—but they are not like our language faculty, which has the defect of ambiguity built right into its architecture. So the question must arise as to how such a defective biological trait originated and why it has not been improved upon over time. One would think there was some selection pressure against rampant ambiguity—that it would have been remedied over time. Yet there is no reason to believe that human language is moving towards less ambiguity, lexical or syntactic. It seems content to remain stuck in its current lamentably ambiguous condition. This is a puzzle: why does ambiguity exist, especially on such a large scale, and why does it persist? It looks like a design flaw of major proportions, so why is it biologically so entrenched? Why don’t we speak unambiguous languages? Why do constructions like “flying planes” exist at all? It is doubtful that comparable ambiguities afflict the languages (communication systems) of other species such as bees, birds, whales, and dolphins; and it would be bad if that were the case given that such languages are crucial to survival. So why does our species settle for anything so rickety and unreliable?
First we must recognize that this is a genuine puzzle—it really is strange that human language is so riddled with the defect in question. Why isn’t there a simple one-one pairing between sign and meaning? Why is the connection between sound and sense so loose? Let me compare linguistic ambiguity with what are called ambiguous figures, the kind found in psychology textbooks (e.g. the Necker cube or the duck-rabbit). These are aptly described as cases in which a given physical stimulus can be interpreted in two different ways—hence “ambiguous”. So isn’t the problem of ambiguity found outside the case of language, and isn’t it really not that much of a problem? But these cases are relatively rare and confined: they are generated by psychologists drawing sketchy pictures on pieces of paper. Seldom do we find anything comparable in nature: it is not as if vision by itself is constantly generating such ambiguities.  We might wonder whether a patch of shade yonder is a black cat or a shadow, but such cases are not common and don’t generally disrupt the purpose of vision. It is not that vision is biologically constructed so as to lead to such uncertainties of interpretation. But in the case of language the problem is endemic and structural: ambiguity is both common and practically consequential. If vision were as prone to ambiguity as language, we would find ourselves in trouble (imagine 32 ways to see a snake, most of them not as of a snake). Ambiguity in vision is sometimes a problem, but it is not ubiquitous enough to thwart the purpose of vision (i.e. gathering accurate information about the environment); if it were, we would expect natural selection to do its winnowing work. But ambiguity in language really is a practical problem, as well as an inherent design flaw: it cuts at the very heart of communication. The question, “What did she mean?” can be pressing and momentous. And the reason for ambiguity in vision is obvious enough: vision is an interpretative, hypothesis-generating process, proceeding from an often-exiguous basis in the stimulus environment, so it must sometimes boldly venture alternative hypotheses. But language has ambiguity built into its syntax, its rules of sentence formation. It is constitutionally ambiguous.
Sometimes a biological trait has a defect as an inevitable side effect of an adaptive characteristic. Thus it is with the human bipedal gait and large brain, or the giraffe’s elongated neck—there is a price to pay for the benefits conferred (in fact, this is true for all traits given that they all require nutritional upkeep). We can see this principle in operation in the case of those ambiguous figures—ambiguity as the price of inference. So could it be that the ambiguity of natural language results inexorably from some super-advantageous design feature? Suppose it resulted from the property of infinite productivity: you can only have that brilliant property if you also have some concomitant ambiguity. The trouble with this suggestion is that there is no obvious move from productivity to ambiguity—why should the former entail the latter? Mere combinatorial structure also doesn’t lead to ambiguity. Artificial languages are productive and combinatorial, but they don’t contain ambiguity. Nor can hypothesis generation be the explanation: true, we have to infer what someone means from the words he utters (along with context), but it is the words themselves that bear an ambiguous relation to meaning. The question is why language permits constructions like “Flying planes can be dangerous” to begin with. Why not just have the sentences, “Flying in planes can be dangerous” and “Planes in flight can be dangerous” (though the former sentence admits the reading, “Flying around inside of planes can be dangerous”)? The fact that the hearer is engaged in an inferential task doesn’t explain why ambiguity of this kind is so rampant and inbuilt. So it is hard to see how it could be a by-product of some desirable design feature; it looks like the product itself. If we just consider quantifier scope ambiguities, we see how inherent to natural languages ambiguity is—and that it is easily removable by some device equivalent to bracketing. So why does our language faculty tolerate it? Why not just clean up the mess? 
One possible explanation is that human language is so spectacular an adaptation that it can afford many rough edges and design failures (compare early wheels). It is so good that it can afford to harbor some vices—it’s still better than having no language at all. This may be backed up by the observation that human language is a recently evolved trait still going through its awkward adolescent phase—eventually it will mature into something more streamlined and fit for unqualified celebration. There may be something to this point, but it doesn’t remove all aspects of the puzzle, because it doesn’t tell us why language evolved with this defect to begin with when better options were in principle available, and there seems no evidence of any movement away from ambiguity heretofore. It might be suggested that ambiguity is like vagueness: it’s not a good thing, to be sure, but tolerable when the alternative is no language at all. I won’t consider this kind of answer further here, because it is difficult to evaluate without further evidence; but perhaps mentioning it serves to highlight the lengths we would need to go to in order to find an answer to our question. I don’t think we would find it plausible that the reason for intermittent blindness is that it is better to have occasionally blind eyes than none at all, but that is essentially what is being proposed by the explanation suggested—communicating by language is such a marvelous gift that serious defects in it can be lazily overlooked by the evolutionary process. Ambiguity is really not like the retinal blind spot. What if our language faculty enabled us to parse and understand only half of what is said? That would be rightly regarded as a grave defect, for which there is no obvious explanation. But ambiguity is rather like that—it really does impede successful communication. And even when it doesn’t, there have to be mechanisms and strategies that enable us to avoid its snares—it’s always less effort to understand an unambiguous sentence than an ambiguous one. Processing speech is certainly not aided by ambiguity. It’s not a blessing in disguise.
It is an interesting question where human language will be in the distant future. Will its present level of ambiguity survive or will it become more perspicuous? Are we now placed on a linguistic path that cannot be altered? What would it take to impose selective pressures on the ambiguity-producing structure of our grammar? At present we have an evolved capacity that tolerates rampant ambiguity, yet functioning well enough to get by in normal conditions; but the architecture is fundamentally unsound, allowing for forms of words that could have many meanings apart from the one intended. Language should make things easier for the speaker and the hearer than it now does. 
 I have found only one paper dealing with the question addressed here (though I am by no means expert in the linguistics literature): “The Puzzle of Ambiguity”, by Wasow, Perfors, and Beaver. There is no date on the paper or place of publication (I found it on the internet), and to judge from its content the problem it discusses is not generally recognized. I welcome any information about other published work on the subject.
 My personal favorite is “Right is right”. Given that “right” can mean a direction, correctness, and moral rectitude, we already have nine different readings of this simple sentence. Now add in the ambiguity of “is” as between predication, identity, and composition and we generate many more meanings.
 I won’t discuss whether ambiguity exists in the underlying innate language prior to its expression in a particular sensory-motor format. It may be that all ambiguity exists at the level of spoken speech and results from the demands imposed by this medium; there may be no ambiguity at the more abstract level of universal grammar. I remain agnostic on this question.
 It might be thought that ambiguity is useful as a means of concealment: say something that some people will take in one way while conveying a different message to others. But this is not a good explanation for why human languages are ambiguous in just the ways they are. It is grammatical rules themselves that allow sentences to be both grammatical and ambiguous, not pragmatic considerations of the kind just mentioned. Ambiguity surely didn’t evolve as a means of selective deception.
 Philosophers are apt to speak of natural language as logically imperfect, as judged by the standards of some ideal language; but ambiguity makes natural language biologically imperfect, because the biological function of speech is communication and ambiguity gets in the way of that. Imagine if a given monkey cry could mean either “Predator nearby” or “Food in the offing”! Compare “Get thee to the bank!” said by someone in the vicinity of both a river and a lending institution.