Cereb Cortex. 2005 Aug;15(8):1261-9. Epub 2005 Jan 5. 

The neural mechanisms of speech comprehension: fMRI studies of semantic ambiguity.

Rodd JM, Davis MH, Johnsrude IS.

Department of Psychology, University College London, UK.

A number of regions of the temporal and frontal lobes are known to be important for spoken language comprehension, yet we do not have a clear understanding of their functional role(s). In particular, there is considerable disagreement about which brain regions are involved in the semantic aspects of comprehension. Two functional magnetic resonance studies use the phenomenon of semantic ambiguity to identify regions within the fronto-temporal language network that subserve the semantic aspects of spoken language comprehension. Volunteers heard sentences containing ambiguous words (e.g. ‘the shell was fired towards the tank’) and well-matched low-ambiguity sentences (e.g. ‘her secrets were written in her diary’). Although these sentences have similar acoustic, phonological, syntactic and prosodic properties (and were rated as being equally natural), the high-ambiguity sentences require additional processing by those brain regions involved in activating and selecting contextually appropriate word meanings. The ambiguity in these sentences goes largely unnoticed, and yet high-ambiguity sentences produced increased signal in left posterior inferior temporal cortex and inferior frontal gyri bilaterally. Given the ubiquity of semantic ambiguity, we conclude that these brain regions form an important part of the network that is involved in computing the meaning of spoken sentences. (My emphasis.)


Here we may have a possible biological locus for exactly the sort of phenomenon I was positing in my previous post. Interestingly enough, ambiguity seems to a core process, and again we have evidence that language users are able to actively engage with ambiguous language and that an important step in cognition is pre-disambiguated. Importantly, it is in all likelihood that linguistic comprehension engages in parallel visualization of multiple possibilities. This is probably responsible for so much of what makes poetry interesting and road signs uninteresting.

The inferior temporal cortex is a higher-level part of the ventral stream of the visual processing system of the human brain. The ventral stream engages in classification and identification of phenomena. The adjacent inferior frontal gyrus coontains Broadmans Areas 44 and 45, which contain a number of non-visual areas heavily engaged in linguistic understanding. Broca’s Area is contained in Broadmans Area 44. Broca’s area is connected to Wernicke’s area via the arculate fasciculus.

One way to disprove my present theory is to see the neural precursors to these differentiated brain areas in fetal development. Do human brains develop the visual system first? Do these linguistic areas develop out of the visual tissues? Or do they come out of a wholly different set of neural tissues? Anyone know a neuroembryologist?

When reading some Steven Pinker a couple of years back I wondered whether language could be better understood via sound, sentence, and vision rather than by words and rules as Pinker suggests (see his Words and Rules). Rules seem to be elements of narration we use or rather abuse to divine a neat model of causality. However there seems to be very little in biology that’s rather rule-like. Biology is inherently anti-functional, at least in the strict mathematical sense of the word function. Cells and subcellular systems can and do appear to regularly do different things given the same input. And that’s assuming we can even truly tightly control an input to a biological system in any meaningful (re: in vivo) way. Weak and strong AI proponents would have us think that neurons are analogues for computer circuits, but the complexity of neural matter is hardly reducible to such a model without sacrificing crucial information.

Rules just don’t seem inherent to language. Words, however, do seem on some level fundamental to language. From a textual perspective certainly. We can see evidence for this in many ways; in my experience the evidence is in building representations of document collections for various text mining experiments. But from an oral perspective, are words fundamental?

Spoken language seems far more continuous that written language not only from a processing standpoint but also from a sensory point of view. Spoken language is experienced and performed in a rather continuous way; words are deduced in learning language, but it remains to be shown whether words are in and of themselves mere narrative convenience for explaining how we understand language rather than language itself. these sounds continue rather fluidly within sentences. The auditory experience of language is that the most coarse break, the most distinct break, is the break between sentences. But spoken language is not just continuous in the way it is serially composed and experienced in an auditory fashion. It is also continuous in that it speads across the sensory spectrum, from sound to vision. Inflection and gesture are essential to processing meaning, and such experience and interpretation is so incredibly integrated and automatic it operates as intuition does.

While the fundamental descriptive unit of language seems to be the word, with the description generating itself through the appearances of language acquisition, the fundamental unit of language seems to be the sequence of sounds, the sentence. The word “book” or for that matter the sound of the word has some basic meaning but no real rich semantics. What book? What’s it doing? Where is it? What’s in it? How thick is it? Do you even mean a thing with pages? Frankly we have no idea what questions even make sense to ask in the first place. The word and the sound alike seem devoid of context, seem completely empty of a single thought. But once we launch into a sentence, the book comes to life, to at least a bare minimum of utility, representation with correspondence to some reality. It seems the sentence is the first level at which language has information.

But it seems that the sentence, the meaning-melody of distinct thought, is composed more essentially with some visual representational content, something rudimentary that is pre-experiential (children blind from birth seem to have no profound barriers to becoming healthy and fully literate adult language users). There seems to be something visual that is degenerative in nature involved in language. Not generative. It seems that language comprehension is based on breaking down the continuous auditory signal into something very roughly visual and then the utterance becomes informative.

My take on such a process is really not so unusual but rather fundamental to one of the most important linguistic discoveries of the modern era. Wernicke believed that the input to both language comprehension and language production systems was the “auditory word image.”

So here’s what I’m thinking. Language’s syntax is not fundamentally linguistic per se nor compositional but rather sensory (audio-visual) and decompositional. So I wonder, is there some sort of syntax for vision, some decompositional apparatus? Or are we just getting back into rule-sets?

I think we can understand something fundamental in this syntax between the sensory and the linguistic. Linguistic decompositon, which is really either auditory or visual decompositon, becomes visual composition in understanding. Likewise, the visual must be decomposed before it can be composed into a sentence.

In other words, if we knew rules for visual decompositon we could automatically compose descriptions of scenes. Likewise we should be able to compose images from decomposition of linguistic signals.

And how do we do that without rules or functions?

But language is not pure sign, it is also a thing. This exteriority -word as object rather than sense- is an irreducible element within the signifying scene. Language is tied to voice, to typeface, to bitmaps on a screen, to materiality. But graphic traces, visualizations are irreducible to words. Their interpretation is never fully controllable by the writing scientist.

– Timothy Lenoir and Hans Ulrich Gumbrecht,
from the introduction to the Writing Science series

I am hung up on a concern about the application of text mining to scientific discovery from which I seem unable to shake free. That simple hang-up is due to the importance of visual analogy to scientific discovery and the rather trivial or secondary narration that follows it. That narrative content (see narrative fallacy – explaining an event post hoc so that it will seem to have a cause) is the very material that text mining seeks to leverage. Language is supposed to capture in some way the network of causes, many of them supposedly sufficient to help presage novel treatments, procedures, further explanations, and so on. But if the generative seed of discovery is visual analogy itself, no amount of linguistic-based reasoning, whether contextual, deductive, or inductive, can ever make new discoveries. Because the explanation is not equivalent to the image.

And yet. And yet we know that we can make discoveries by deductions from multiple texts, as Don Swanson has repeatedly shown us. But Swanson’s discoveries using disjoint literatures are marginal and hypothetical and remain in desperate need of empirical review. Disjoint literatures don’t appear to be radically increasing the speed at which scientific discovery is made, which means that the process of leveraging implicit multi-document logics is missing something essential.

I’ll venture a guess and say that pictures are missing.

If a picture is worth a thousand words, is the relation symmetric? That is to ask, given a thousand words, can we draw a picture? Could we, say, use hypothesis generation to augment the creation of visual metaphor apparently crucial to scientific discovery? Alternately, it seems that a picture is not inherently worth any word whatsoever, and that inequivalence is symmetric.

Most pictures generated these days via automated means are entirely dimensionless, metaphorically speaking. Graphs, trees, constellations of points in a space. But what makes our understanding of constellations rich? Ahh yes, those stars in our southern summer sky appear to look like a scorpion, become known as Scorpius, and that’s how we remember those specks, and that’s how we use them as well. Memory, after all, is inseparable from use. And yet those stars are no more a scorpion than a snake or a lock of hair or whatever else you can make up.

So it’s not enough perhaps to plot networks on a 2D screen. Why not compare those assemblages of seemingly random points to visual shapes? Why not revtrieve the visual metaphor for an item automatically?

This however is utterly unconvincing. There’s no way, for example, special relativity could be arrived at in such a way. And yet, hold on just a sec, elements of the discovery of special relativity are in part a result of a visual search activity–Einstein imagining many rich ways of illustrating previous mathematical expressions and testing the illustrations to measure their utility, their usability, their ability to survive multiple looks and provide a rich metaphor capturing the scientific phenomenon. And then using those images to tell further stories, and then usiong those stories to generate more mathematical expressions. A picture is worth a thousand words and a thousand words is worth many pictures.

For my master’s thesis I performed a case study of a very large multinational drug company to evaluate how it innovates in text mining to drive its central mission of drug innovation. Drug discovery is hard and therefore expensive, but with high performance computing now a commodity, drug companies should be at the bleeding edge of text mining innovation, particularly in the area of virtual hypothesis formation and testing (deriving novel insights from mining multiple inputs, from clinical data stores to genetics databases to research literature collections and even the so-called grey literature). But guess what? At least with respect to the case I studied, they aren’t. They are highly focused on circa-1997 extraction tasks with little to no interest in statistical learning and a confused interest in taxonomies and automated inductive reasoning. They invest in formal logics and in information extraction but the meat in the middle, the statistical learning, is kept strictly to data mining of data sets severly limited in scope. Simply put, the company has little to no coherent and well-articulated vision of how it can tackle its most daunting problem for drug discovery: information overload.

How can this problem arise? Isn’t the central mission of a drug company, its core competency, to create new drugs? Well, historically it has been. But competing with the core competency of the drug company is another, oft conflicting, central mission, to make money. What this means for drug discovery is that it is only kind of important. The company I studied was laying off key drug innovators globally as it was focusing its investments further down the drug pipeline, placing more and more emphasis on Phase 2 & 3 projects, more on lower risk short term gains. What this means is that the central mission has become, to get drugs to market, particularly ones with a recurring revenue model.

Historically the drug companies could hang their hats on introducing drug treatments that have contributed to huge improvements in human health over the last century. Drug companies have been in the business of saving lives. Drugs are largely responsible for the 50%+ increase in life expectancy in the US over the last century.

Sometimes, however, human health improvements are not profitable. Sometimes drug companies will select strategies far less beneficial to human health that are far more financially beneficial to the organization. Consider the focus on marketing deregulation in the US, or FDA deregulation. Why invest in developing drugs when you can invest in removing barriers to sales? Now that deregulation has just about all but run its course, drug companies will soon face the fact that they will need to depend more and more on releasing new drugs. When the two largest drug companies in the world can’t combine for more than a dozen new drugs in any given calendar year, you can tell that something’s clearly broke. You can’t hang the shortage on regulation or on a shortage of actionable research.

So why the institutional emphasis away from innovation? One can only speculate; I will use Portfolio Theory to speculate. The dominant forces controlling large multinational drug companies are people of a certain kind, namely, aging investors. They have invested their dollars and expect something in return. Portfolo Theory tells us that our optimal investment trend as we age is to go from high risk to lower risk, income-generating investment. For example, I’m 35, and if I expect to live to, say, 80, I’m probably at least three decades from retirement, from a time where I need my investments to generate income. Because I have decades to invest, I can handle the risk of higher risk investments, namely because I don’t need the reliable income, and because I have time to recover if I lose. The game of investing depends entirely on how much time you perceive yourself as having, namely because on your death bed all the money in the world is worth nothing, but having a lot of cash on hand that last week before your death bed is pretty damned important. A promise for a check next week won’t do you good if you’re dead. And so I think, unlike me, the investors in large mutinationals are old men, frankly. They need to allocate their assets on the income-generating end of the spectrum. They need that cash and they need it now.

This asset allocation model is confirmed by the reduced interest in technological innovation and the increase of interest in being merely early adopters. Adopting established technologies carries a lower risk, as it has a higher probability of some payoff.

And so why invest in high risk, in innovation? The argument for it would be three-fold: to attract younger investors, to focus on the longevity and long-term stability of the company, and to be true to the core mission, which should be to treat health ailments. Maybe my experience is anecdotal, but at least to me it appears that investors in my parent’s generation (they’re 65) are far more likely to invest in, say Pfizer or GSK, than investors my age. They’re safe, they’ll do OK next quarter, but that picture is very murky a decade from now. Not to mention that younger investors no longer see drug companies as beneficial to human health. There’s nothing attractive for the average younger investor.

One of the saddest consequences of this reluctance to innovate, this focus on profit, is the impact on human health. Drug companies are far more willing to repackage old drugs and market the heck out of them, renewing their proprietary charges, than to find new drugs. And when the drug companies choose new drugs to invest in, they are going to look for “comeback” drugs, drugs that cure nothing but treat indefinitely. No new antibiotics are reaching market because there’s no incentive

I haven’t posted in six weeks.  It’s not because I have nothing to write about.  Rather it’s that I have been extraordinarily busy, mostly in my search for employment.  Fortunately and I do mean fortunately it looks as if the search is drawing to a close.  I’m ready to put my shoulder to the wheel for the right people, the right project.  It’s much harder to find a job in numerous ways when trying to be selective.  It is apparent to me now that, in addition to testing my fortitude, it will pay in the long run.  I hope to share more information as my hunt officially comes to an official resolution.  Namely because I am bubbling with excitement over it.

What I can say is, is that the future is here.  I mean really here.  I mean it in more ways than one.  As in, the technology of the future is now here, and it can bring more of the future into the present.  That means the future is *really* here.

The following comprises a collection of my intuitions and “big picture” insights resulting from graduate study focused on text mining at SILS. These are insights related to feature representation, knowledge engineering, model building, the application of statistics to real-life phenomena, and the greater whole of information science.

Many of these apparently go without saying, yet so many discussions of supposed problems would go away if some of these observations were made explicit. This is my attempt to make them explicit. Maybe it goes without saying that expressing the obvious is sometimes quite necessary.

1. Statistical models often fail because they’re missing key attributes necessary to describe the phenomena they represent

Attributes that are altogether unrecognized, difficult to quantify, difficult to analyze, truncated out, or simply forgotten arguably dominate and confound the predictive/explanatory power of statistical models. These missing variable abound. Their absence dominates to the point where theory itself must give way to empiricism and its sister, skepticism. It also means that we simply don’t see everything and that it never hurts to try and see more things.
2. Feature reduction of highly dimensional linguistic data sets is a misguided, outdated and counterproductive approach

There. I said it.

Claude Shannon’s model of information as that which is located among noise is a metaphor that appears to have been misleading a number of people in information science, particularly those involved with anything even remotely tangential to text mining (or, if you must, “knowledge discovery”). Information in an atomic form (e.g., bits) allows for the differentiation of signal and noise. A bit either is a signal or it isn’t. Attributes of real-life phenomena (e.g., average first down yardage in football for a team) are not like bits, at least not in the way we experience them and interpret those phenomena, whether in written explanations or in databases. “Real-life” phenomenae comprise different sorts of real-world features that can never be honestly reduced to their atomic constituents. And, pragmatically speaking, they won’t be reduced to quantum atomic states any time soon.

Given that every attribute of real-world phenomenae we identify partakes of both signal and noise, the removal of any attribute (save for the case of redundancy) always corresponds to the loss of information. Ultimately the statistical modeling of phenomena such as competitive sports and stock markets and clinical emergency room chief complaints is wholly unlike modeling communication channels. There’s something immediately discontinuous about binary electronic signals while other these other phenomena need dramatic interpretive steps before they can be represented with discontinuous electronic signals. Finally, signal and noise are terms that don’t apply very well because that which we are modeling can only be realistically described by features that are both informative and misleading at the same time.

There’s something rather continuous about language (something that latent semantic indexing attempts to capture) and that even the simplest of approaches, such as applying stop word lists to bag of word representations, lost critical information that dictate the semantics of the document. “Dog,” “a dog” and “the dog” quite clearly mean different things, as do “of the dog”, “out of a dog” and so forth. Representing all of those quotations as “dog” or going a step further and representing all of these quotes with the very same word-sense identifier, dumbs down human language beyond recognition. Garbage in, garbage out is a phrase I learned more than a quarter century ago when learning to program games for the Commodore Vic-20.

Reading a text book from 1993 on the C4.5 algorithm, I came across reflections that some crucial elements of C4.5 appeared to be motivated by economizing on computer resource issues. Not enough memory, too slow of processing, etc. In 2007 high performance computing is a commodity. The pressures for feature reduction in machine learning needed to be heeded 14 years ago, but they’re considerably less of an issue today.

Finally, at the very end of my stretch of graduate school studies I accidentally came across a new strategy for feature representation that is so painfully obvious in retrospect it leaves me wondering why no one else has been doing this. Fortunately for Hypothia it spells one very big competitive advantage. But I digress.

3. There’s always something missing from your set of attributes (cf. 1 & 2)

4. There’s no substitute for knowing your data set (cf. 1)I credit this oft-neglected, oft-devalued approach to my first and truly excellent data mining instructor, Miles Efron, who may be to blame for turning me on to text mining in the first place. What have you wrought? He made sure to repeat this lesson of knowing thy data a few times, and the lesson was surely not lost on me. In fact it seems as it it frames and justifies my confidence in my approach.

5. [DELETED] and let your algorithms optimize your attributes for maximal classification margin (cf. 2 & 3)

Can’t say the deleted part yet. But I will, eventually. It probably should be obvious by now. But still I’m not prepared to say.

6. SVM+SMO is very good for binary classification of highly dimensional data (cf. 5)

Improvements to SVM+SMO are always welcome of course, and it appears there are now numerous implementations of SVM that improve. I should note that, according to Eibe Frank, SMO in Weka (written in Java) is just as fast as Joachims’ SVM-light written in C. SMO’s pretty good.

SMO solves the QP problem created by SVM efficiently.

7. You always need more computing power (cf. 2, 5 & 6)

The curse is not dimensionality, the curse is not intellectual. The curse is economic, a problem of resources.
Likely it will be difficult to produce a dataset that is intractable for a good HPC setup running SVM+SMO but it doesn’t exactly hurt to try as long as you’re trying to harness more and more power.

8. You don’t know everything (cf. 3 & 4)

9. models only forecast well in forecast-influenced environments only when the model has an information advantage over other models (information assymetry, competitive advantage)

10. You’ll never get it quite right ( cf. 8 )

11. There always more left to do (cf. 5, 7 & 10)

12. Disambiguation can be better pursued not in any pure sense by machinic strategies but rather by messier approach of utilizing the greater context surrounding term, document, and corpus, which in turns permits some degree of ambiguity, which is necessary for understanding

13. Word sense disambiguation is quite possibly the wrong way to go to conjure semantics in one’s text representation (cf. 2 & 12)

As I’ve written before, there are other approaches available to leverage semantic information that are better than word-sense diambiguation (WSD) .

14. More formally, the incorporation of ambiguity into linguistic representations (i.e, representing all possible word senses/meanings and POSs for any given word) allows for better representations of intelligence than ones produced at least in part through WSD strategies

15. For artificial intelligence to become smarter than humans, it must at least be as smart as humans first.  A person’s ability to understand multiple senses of a given word at once (of which poetry is perhaps the most striking example) is strikingly intelligent and far more intelligent than most WSD approaches I’ve seen (cf. 14).  And when you consider that the basic unit of meaning is truly not the word but the sentence, WSD seems all the more foolish, and yet makes me feel there’s a huge opportunity to understand language from its wholes and holes.  Discourse analysis anyone?

16. Not knowing everything, not always getting it right, and always having more left to do makes the hard work a great deal of fun. Discoveries are everywhere waiting to be written into existence. (cf. 8, 10, 11)

17. Don’t panic, be good, and have fun. (cf. 16)

18. The essence of human language is nothing less than the totality of the human language in all of its past present and future configurations and possibilities.