Of Cheese Making Computer Overlords

Next time somebody tells you that you should never
use multiple inheritance,
look them straight in the eye and say,
"One size does not fit all."
If they respond with something about
their bad experience on their project,
look them in the eye and repeat,
slower this time,
"One size does not fit all."

Marshall Cline, C++FAQ Lite

In a classic example of Baader-Meinhof, having never before heard the word 'singularity', I suddenly found myself hearing about it all the time. Initially, it was a friend interested in the technological kind who introduced me to the concept, but soon thereafter it began to mushroom in conversation everywhere: black holes, limits, power laws. Suddenly the name of the film `Event Horizon` made a lot more sense.

It happened, then, that this inconsequential realisation was idly forming in my mind one day just as I was reading about the hidden complexity of wishes. So what of it? In the article the author argues a non-sapient genie could never be trusted to grant you a wish, since to make the correct choice it would have to understand all of its non-explicit suppositions, ones which you yourself may not even be aware of. One moment you tell your genie 'quickly, get my Mother out of the burning building', the next her disembodied bits fly past you as a boiler explosion propels her far away from the center of the house: the quickest way of getting her out, but perhaps not what you had in mind. Somehow this reminded me of Google image search.

As human beings we're quite a bit better at making choices than our non-sapient silicon friends, but we are still far from perfect. Take the realm of justice. Were we to act as the wish-granting genie of a man who was asking for a book of good laws - how could we respond? Informed by a history of moral philosophy, we might advise the man that any decisions he makes must be consistent, that they must treat everyone as equal unless there are relevant reasons for treating them otherwise, in which case the unequal treatment should be proportional to its cause. But this is hardly very helpful.

Using a parable from a book by the philosopher David Miller, imagine the man had been given £100 to allocate between five people. The rules given so far would instruct him to treat them equally; or differently, but only to do so for relevant reasons and in proportion to them. Thus, the five men might be his employees and the £100 their weekly bonus: then, he should reward each based on the contribution made to the joint enterprise. If he is an aid worker and the five men members of a destitute group, he should look at the urgency of their needs and split the money accordingly. Or the £100 may be the prize to an essay competition, in which case all of it should go to the author of the best piece. It may also be a lottery win, with the six men part of a syndicate which required the sum to be split equally.

In other words, despite having some idea of the general principles involved, a just decision must be highly contextual. The rules are few and the exceptions many. For instance, we may say the principle of equal treatment is paramount. But in a given context, need, right deserts, promising and contracting, restitution and compensation are all things which trump it.

The lack of context is what will make the genie propel your Mother out of the burning building and what makes computers hopelessly bad at making intelligent decisions. What our patron above had wished for was not the wisdom of doing the right thing in this or this other particular circumstance, but a theory of justice - a formal specification of how to resolve any situation. Of course, had we had that, we would had achieved a sort of ethical singularity. We could program a computer to do it. But we can't.

The Stuff of Thought

If olive oil is made from olives, what do they make baby oil from?

A writer is someone who writes.

But fingers don't fing, grocers don't groce,

hammers don't ham and haberdashers don't haberdash.

If the plural of tooth be teeth, should the plural of booth be beeth?

One goose, two geese - so one mouse two meese?

If a man wrote a letter, perhaps he also bote his tongue.

Richard Lederer,  Crazy English

Why do we have noses that run and feet that smell?

Random Person on the Internet

How much money should the insurer have paid out to Larry Silverstein, the leaseholder of the World Trade Centre site? Silverstein held insurance policies that stipulated a maximum reimbursement for each destructive `event`. So was he due $7bn (two towers, two collapses, two events), as his lawyers argued, or was he due only $3.5bn (one plot, one event)? With this unusually expensive debate over semantics, I was introduced to Steven Pinker's book on `The Stuff Of Thought`*.

If you are a native English speaker, you'll probably have no trouble picking the correct sentence from the choices offered below:

• Tex nailed posters onto the board.
• Tex nailed the board with posters.*

• Serena coiled a rope around the pole.
• Serena coiled the pole with a rope.*

• Ellie covered the bed with an afghan.
• Ellie covered an afghan onto the bed.*

• Jimmy drenched his jacket with beer.
• Jimmy drenched beer into his jacket.*

• Hal loaded the wagon with hay.
• Hal loaded hay into the wagon.

But how do you do it?! And why does one of the sentences in the pair sound so odd? Since both phrases are equally as successful at conveying the meaning (would you ever be puzzled by someone saying Amy poured the glass with water ?), what is it that makes one of them wrong? Why can't all verbs work in both a content-locative construction (which is the formal name of phrases where the content is the focus) and a container-locative one (where it's the container that sits under the spotlight)?

• Hal loaded the wagon with hay.    < --- container-locative (A)
• Hal loaded hay into the wagon.    < --- content-locative (B)

How in fact do speakers (and, more intriguingly, small children) work out that certain verbs can't appear in otherwise perfectly good constructions? Or work out that certain other verbs can? (Notice that in the last pair, both sentences are equally valid). Here are then two formal questions:

1. Question #1: Can one teach a computer to reliably generate a content-locative construction given a container-locative one? (And vice-versa?)
2. Question #2: Can one identify the property of verbs which designates them to the two camps - those that work in both constructions (as in load) and those that don't (as in drench)?

Virgin Mary on a Cheese Sandwich?

Often, seemingly haphazard linguistic patterns such as the above are in fact underwritten by a formal rule. For instance, the mystery of why -er and -est can't be applied to certain adjectives (you can't be specialer that your friend or the beautifulest in the world) found its match when someone noticed they can only work with words that are monosyllabic (redder, nicer, older) or have at most an insubstantial second syllable (prettier, simpler, narrower).

The oddity concerning the verbs above (with some that work only in content-locative constructions and other only in container-locative ones) puzzled linguists for some time. This was, in part, due to the influence of computer scientist-cum-linguist Noam Chomsky, whose work made people inclined to think of language rules as operations of cut-and-paste. Saying that the locative rule concerning contents and containers cared about the meaning of the verb was like claiming that 'your word-processing program refused to cut and paste words with certain meanings while happily doing so with all the other ones'.

But as it turns out, meaning matters. The content-locative constructions can be thought of as saying "A causes B to go to C" (Hal loaded hay into the wagon.) while the container-locative that "A causes C to change state (by means of causing B to go to C)" (Hal loaded the wagon with hay.) These two ways of thinking about the same event are a simple and elegant explanation of why the locative rule allows certain verbs to take on the first construction, others the second and others still both of them:

A rule of semantic reconstrual (the gestalt shift): If a verb means "A causes B to move to C" it can also mean "A causes C to change state by means of moving B to it". In both cases, the affected entity (B for meaning #1, C for meaning #2) should be expressed as the direct object.

Now, words in Hal loaded hay into the wagon and Hal loaded the wagon with hay don't just move around for no particular reason. The rule says the affected entity must go from oblique object to direct object. Still - the rule so far only answers Question #1. What about why some verbs can make the transition while others cannot?

Stick Men

To answer that, the first thing you must notice about the pairs of sentences above is that the two formulations are not perfectly synonymous. When one loads hay into a wagon it can be any amount (even a couple of pitchforks). If, on the other hand, the wagon is loaded with hay then we might reasonably expect it to be full once the action has been completed. Linguists call this subtle difference the holism effect.

The holism effect is not a mere accident, but rather follows from the very nature and purpose of the locative rule: to change the focus from the content to the container or the other way around. And it reveals something about how the mind conceives such a change. In fact, the holism effect applies to direct objects in general**:

• Moondog drank from the glass of beer where the glass is an oblique object of from, and is consistent with him taking only a few sips, while
• Moondog drank the glass of beer where the glass is now a direct object, means he's drunk all of it.

A Direct Object is a pretty important syntactic role (second only to the Subject and Predicate). The glass is more in the mind's forefront in the second sentence then in the first. In the examples above, the affected entity is always the Direct Object, and doesn't quite make it to Subject rank only because an agent is also present in the sentence (and we care more about agents than we do about objects, perhaps for evolutionary reasons, since agents can eat you while objects tend to be more passive***).

In fact, sometimes the holism of the affected entity (the Direct Object) rubs off onto the agent herself (the Subject), as if the mind is presented with 'a sensuous image of an entity so saturated with stuff or bits that the mind blurs the two and apprehends the entity as doing what the stuff or bits ordinarily do':

• Bees are swarming in the garden.
• The garden is swarming with bees.

• Juice dripped from the peach.
• The peach dripped with juice.

The reason this is even possible is because the English language treats a changing entity (usually the Direct Object) in the same way that it treats a moving one (usually the Subject). This for instance doesn't happen in Romanian****. Ray Jackendoff, a linguist, explored much of how words denoting motion in a physical space often come to denote change (a kind of metaphorical motion) in a state-space, as in

• The manager kept Petro at work.
• The doctor kept Pedro well.

• Chris went to Paris.
• Chris went crazy.

We, in fact, use metaphorical translations of space concepts all over the place (we say stock markets have risen and that our mood went down). Good is up and bad is down, the future is in front of us and the past behind and events on a time scale travel towards us (the end of the night drew nearer) or away from us (the meeting was moved back to Monday) - though they can also stand still and wait for us to come to them (we were approaching January).

Furthermore, when we conceptualise an entity moving in state-space we apply the same schematization to it as we do to things moving in the physical world. This means that 'when the mind conceptualises an entity in a location or in motion it tends to ignore the internal geometry of the object and treat it as a dimensionless point'. And since we say a wagon goes from empty to full the same way we say a car has gone from London to Paris, it now makes perfect sense that the wagon does not allow intermediate states in the container-locative construction: spatially we tend to think of the car as either in London or in Paris, not giving much thought to the road in between.

Finally we now have all the ingredients required to answer Question  #2: that it is both the meaning of the verb as well as the meaning of the construction which inform us of which verbs can accept the locative alternation. This allows us to understand why we can Throw a cat into the room but cannot Throw the room with a cat - since we cannot conceptualise throwing something into a room as enough to change the state of that room.

Another example. To pour means to allow liquid to move from A to B in a continuous stream. It specifies a relation (letting not forcing for instance) and the attribute of the motion (slow perhaps). It says something about the motions but nothing at all about the state of the container: you can pour water in a glass (making it full), onto the floor (making it wet), over a person (making her angry), but nothing predictable happens to the receiver. In that way it is different from spew, splash and spray which say intrinsically something about how the state of the container has changed: something can be spewed, splashed or sprayed but not poured. Which is why we can say Hannah poured water into the glass but not Hannah poured the glass with water. Fascinating, isn't it?

* This is a book I warmly recommend - all the examples above are shamelessly cherry-picked from there. Also, you might want to check this guy out on TED, his talks are usually quite entertaining.

** Remember, in the locative rule, first the container was the direct object (Hal loaded the wagon with hay), then the content (Hal loaded the hay into the wagon).

*** Just by the by, when an object becomes active, as in the tree that's about to squish you, or the car that's about to run you over, both linguistically and mentally they become agents which is why we say that we tend to anthropomorphise objects when they do things. As it turns out, this might be one of the reasons why the concept of deities comes so naturally to most of us (since natural phenomena happen, it is a mental reflex to ascribe agent features to them). But more about that in a another post.

**** Albinele bazaie in gradina cannot become Gradina bazaie cu albine.

Control

Lacking the normal, protective barriers of inhibition,
the normal, organically determined boundaries of self,

the (...) ego is subject to a lifelong bombardment.

He is beguiled, assailed, by impulses from within and without,

impulses which are organic and convulsive,

but also personal (or rather pseudo-personal) and seductive.

How will, how can, the ego stand this bombardment? Will identity survive?

Can it develop in face of such a shattering, such pressure - or will it be overwhelmed (...)?

[Can it be] held whole and sovereign or [will it] be taken over,

possessed and dispossessed,

by every immediacy and every impulse.

Oliver Sacks, The Man Who Mistook His Wife For A Hat

Certainly, here Oliver Sacks is talking about Tourette's . But is this not, grotesquely exacerbated, an eerily accurate portrayal of a certain temperamental disposition? A lurid reminder? It is hard not to occasionally remember how the mantra that we belong to ourselves may readily disappear: now a second nature reality, tomorrow a fugitive phantom, the stuff of thought, held in place only by a painfully conscientious process of the rational mind.

Living in a world void of any certainty, a permanent anxiety looms beneath the surface of every moment. All action, thought, decision - however simple and inconsequential - carries the burden of a Damocles' sword. Tomorrow's ego fears the temper and resolutions of today's. And when tomorrow lies permanently under the sharp blade of a potentially radical change of heart - how can the ego not live in a state of permanent anxiety? There is no control. One is merely swept away from one shore to another, where its reason must constantly dock and deal - held in a state of painful, permanent acuteness - with whatever happens to have come its way.

There is a lot to be said about the comfort of feeling at peace with one's own identity - a sense of self that does not shift and metamorphose incessantly, right from underneath the stand of reason. That does not make the ego a stranger to his own mind, stranger still to all of his loved ones. That braves the flow of time. That stays recognisable, giving some support to this illusion of the permanency of the soul. That tomorrow it shall not awake trapped - stripped of its proprioception of the psyche - into the unrecognised, unassumable and sometimes ghastly imprisoning life of a past-self long gone.