Office of Science & Technology - The Unknown Second Nature

The Unknown Second Nature - Introducing Linguist Thomas Graf

bridges vol. 32, December 2011 / News from the Network: Austrian Researchers Abroad

By Carmen Cornelia Lajtos

Thomas Graf

As Christmas approaches, electronic commerce companies like amazon.com are already celebrating the business of the year. But be cautious: A short session of browsing amazon.com may lead not only to books or DVDs, but - in case of Ph.D. candidate Thomas Graf - to a life's work. What started with an innocent order of some "kind of cool looking" introductory book on linguistics, resulted in two master's degrees in linguistics and in starting an academic career at the Department of Linguistics at the University of California, Los Angeles (UCLA).

To Be or Not to Be Smart - That's Not the Question

"Asking a linguist how many languages he speaks is like asking a doctor how many diseases he has," states Graf, who obviously has been asked too many times about the number of languages he speaks. In fact, "linguistics is as much about speaking many languages as mathematics is about calculating with really, really big numbers."

Linguistics is about language. It is about the mental structure and processing of language, and it analyzes how the use of language depends on varying contexts and situations. The topics of linguistic research are present in our everyday life:

“If a child has an actual language impairment, the research done on language acquisition is of great importance in figuring out how to help him or her.” – Thomas Graf

Such experimental linguistic research with children is conducted in laboratories like the Psycholinguistics Laboratory.

Just think of toddlers producing their first words. Within a few years children are able to learn the core grammar of their language effortlessly and without direct instruction. Furthermore, children uniformly undergo the same phases during first-language acquisition independent of the specific language they are exposed to. Based on these and other observations, Noam Chomsky, founder of generative linguistics, introduced his idea of some innate "language acquisition device" in the late 1950s. This "device" is supposed to be activated by linguistic stimuli and enables us to easily learn the grammar of any human language at a very early age. However, since "language is second nature to us," remarks Graf, a proponent of Chomsky's idea, no one is aware of "the complex computations going on in our heads while we are talking." Theories like Chomsky's try to provide explanations for these complex processes involved in acquiring and processing language. In addition, Chomsky's theory claims that the acquisition of grammar is the acquisition of an autonomous cognitive ability. This means that nonverbal cognitive abilities, or what is generally referred to as "intelligence," do not directly influence the development of language acquisition. As Graf puts it: "No matter how smart or 'not-so-smart' you are, if you don't have an actual language impairment, you will be fully capable of verbal communication." In order to establish norms for the "standard" language development of children, and also to determine language impairments, neurologists, psychologists, or linguists like Graf test the linguistics skills of children in laboratories like the Psycholinguistics Laboratory at UCLA.
One concrete peculiarity in early language development that has often been the subject of experimental research is explained by Graf: "If you say, for example, 'John likes him,' then 'him' cannot mean 'John.' It cannot mean 'John likes himself' but 'him' must be some other person. Young children don't really get that: They are OK sometimes with 'John likes him' meaning 'John likes himself.'"

In general, Graf's research is of a more theoretical nature: He works on the elaboration and substantiation of Chomsky's proposed "Minimalist Program" for the mental processing of language. On that point, he spends his time exploring the principles that define which combinations of words are possible in order to end up with grammatical sentences. When asked how he would explain his research, Graf likes to keep his explanation pretty abstract but still easy to follow:

Syntax trees illustrate the assumed structure of an arbitrary sentence. "Linguists assume that a sentence is grammatical if it is possible to assign it a special kind of geometric structure that encodes the dependencies holding between the words in the sentence."

- Thomas Graf (http://tgraf.bol.ucla.edu/research.html)

"Linguistics is a little bit like the chemistry of language. What we do when we're talking is, we're building really complicated structures and the question is 'How does that work?' Metaphorically, the words are our atoms, and when we build sentences we combine them into molecules. And just like in chemistry where some molecules are stable and others are unstable, in language certain sentences are good and some sentences are not good. If I say, for example: 'What did who buy what?' it is not a good sentence; but the question is, why it is not a good sentence. And just as a chemist will have a theory about which molecules are stable and which molecules are unstable and why this is the case, a linguist will try to figure out why some sentences are good and why some are not good."

When asked about a "real world" application of his work, Graf provides an example from the field of machine translation. This discipline is about developing and improving programs that translate text or speech from one natural language to another. Systems like Google Translate use statistical information to create translations. However, such statistical machine translations neglect syntactic structures, for example, which are the focus of Graf's work. His research investigates the syntactic rules that compose sentences in a grammatical way. Graf claims that statistically based systems currently do not take syntactic structure into account and also ignore other linguistic features like context. "If people want to improve the performance of those models they'll have to incorporate more linguistic ideas. And that will be the point where research like mine really matters," states Graf.

Math - the Study of "Funny Symbols"

Explaining linguistics in terms of chemistry is not the only time Graf sets foot on natural science ground. Actually, if one happens to run into him and asks about what he is doing, one should not be surprised if Graf replies: "I'm doing math." This white lie mostly "shuts people off pretty good" and prevents Graf from having to answer questions like: "What the hell is linguistics?" From a certain point of view, this answer is also justifiable, in light of the special subfield of linguistics in which Graf works: mathematical linguistics. Mathematical linguists use mathematical methodologies and tools to obtain and to formalize linguistic results.

Most people do not expect to encounter mathematics in linguistics literature. Thus, they are all the more likely to be puzzled by discovering abstract algebra in linguistics journal papers. At least, Graf had this reaction when he stumbled on some mathematical linguistics article in his third year as a linguistics student: "I had no idea what that was, but I knew that I wanted to understand it." This article, written by Edward Stabler, professor of linguistics at UCLA, was "life changing" for Graf. Therefore, when asked about his motives for leaving the University of Vienna to continue his studies at UCLA, his answer comes quickly: "Well, that's easy - because Ed Stabler was there."

Even if linguistics students have no intention of focusing on this highly mathematical subdiscipline in the way that Graf and his Ph.D. advisor Edward Stabler do, there is no way to get around at least some "basic math." In his position as a teaching assistant at UCLA, Graf has some important advice for all linguistics students to whom set theory, mathematical proofs, or mathematical logic are completely unknown territory: "Don't be afraid of funny symbols."

The following simplified example drawn from semantics, the study of meaning, illustrates the interaction between the mathematical branch of set theory and linguistics. Consider the sentence:

"Tommy is a cheerful Christmas elf."

Semantics tries to provide the conditions under which this sentence is true. Consequently, the goal of the semantic analysis of "Tommy is a cheerful Christmas elf" is to calculate the truth-conditions of that statement.

In order to start, define some set A containing all cheerful individuals in the world and some set B containing all individuals that are Christmas elves (for the sake of this example, suppose that Christmas elves do exist). Now note the intersection of the two sets. The intersection of A and B (abbreviated: A∩B) yields those individuals that are both cheerful and Christmas elves. As a consequence, if and only if Tommy is in the intersection of A and B - which means that Tommy is both cheerful and a Christmas elf - is the sentence "Tommy is a cheerful Christmas elf" true. Otherwise it is false.

It is not immediately obvious why one should put that much effort and calculus into the simple task of determining when "Tommy is a cheerful Christmas elf" is a true statement. However, a profound mathematical theory that provides some stable framework will prove to be more beneficial as sentences become more complicated. As Graf explains it: "I think the perspective that mathematics gives you on all those linguistic issues is incredible. When you just do 'normal' linguistics, you could change this or that part of the theory and it wouldn't be so clear what the consequences would be. If you use math, you will know if the change is just cosmetic or actually really fundamental to your proposal. With math, you just go in there, you prove it, you know it. Done."

It is exactly this reliability that mathematics provides in doing research that equips Graf with enough foothold when he is "out there pushing the boundaries of what we know." When "toying around with ideas" concerning the next step in his current research, Graf often feels like "a pioneer that is settling the west, looking at all the stuff nobody has ever seen before." But this "romantic" perspective on doing scientific work and, especially, Graf's literal "settling the west" when he moved to Los Angeles in 2008, require financial support. Fortunately for Graf, the Department of Linguistics of UCLA has been funding his work ever since his first steps in the "new world," and in 2010 the Austrian Academy of Sciences awarded him its doctoral fellowship of €30,000. Speaking of funding, Graf thinks that applying for grants, scholarships, etc. is "less competitive" in the humanities than in the sciences where more people actually apply for financial support and where more money is needed, for example, for running research laboratories.
Linguistics - the "Rigorous Formal Kid" Among the Humanities

For Graf and other linguists, the classification of linguistics as belonging to the humanities is not obvious because of the great interdisciplinarity of the field. As Graf puts it: "We have our fingers in every pie," indicating their brisk interaction with disciplines like psychology, neuroscience, or social and media sciences, as well as computer sciences or mathematics.

“Basically, if you wanted to place linguistics somewhere on this picture, you couldn’t place it anywhere because we span it all.” – Thomas Graf

According to Graf, the core of linguistics, or especially of theoretical linguistics, should certainly be treated as a natural science. On an organizational level, though, regarding the aforementioned funding, Graf feels that "the consideration of linguistics as part of the humanities is not unfavorable." Furthermore, Graf regards linguistics as a young, formal "offspring" of the humanities that follows a somewhat more abstract path than its "ancestors" and that likes to challenge "newborn" linguists: "I think many undergrads taking their first linguistics course are actually surprised by the level of abstraction. Considering that linguistics is much younger than, say, history or English literature, I guess you could call us the rigorous formal kid among the humanities."

The regular entanglement of linguistic skills and formal, abstract thinking in Graf's special field raises the question: Does one have to be greatly talented in both linguistics and mathematics in order to address this research area? For those wondering about this, Graf concludes: "At the end of the day, all it takes is to have creative ideas and a feeling for how you can make those ideas precise."

Maybe some new "creative ideas" will cross Graf's mind on his flight home to Austria for Christmas. In light of all the effort Graf has already put into his linguistic research and plowing through vast amounts of mathematical literature, let's just hope that there will not be any new "kind of cool looking" books awaiting him beneath the Christmas tree!

***

This article is based on an interview conducted by the author, Carmen Cornelia Lajtos, with linguist Thomas Graf, a Ph.D. candidate in the Department of Linguistics at the University of California, Los Angeles (UCLA).

References:

Graf, Thomas. "Closure Properties of Minimalist Derivation Tree Languages." LACL 2011/Lecture Notes in Artificial Intelligence, edited by Sylvain Pogodalla and Jean-Philippe Prost, 6736: 96-111. Heidelberg: Springer: 2011 <http://tgraf.bol.ucla.edu/doc/papers/LACL2011.pdf> (accessed December 6, 2011).

Graf, Thomas. "Locality and the Complexity of Minimalist Derivation Tree Languages." To appear in Proceedings of FG-2011. 2011 <http://tgraf.bol.ucla.edu/doc/papers/fg2011.pdf> (accessed December 6, 2011).

Graf, Thomas. "Locality and the Complexity of Minimalist Derivation Tree Languages." FG-2011. Ljubljana, Slovenia: August 6, 2011 <http://tgraf.bol.ucla.edu/doc/talks/fg2011_slides.pdf> (accessed December 6, 2011).

Heim, Irene and Angelika Kratzer. Semantics in Generative Grammar. Malden, Mass. [et.al.]: Wiley-Blackwell, 1998.

Psycholinguistics Laboratory. Department of Linguistics. University of California, Los Angeles <http://www.linguistics.ucla.edu/faciliti/psychlab.htm> (accessed December 6, 2011).

Pullum, Geoffrey K. and Andras Kornai. "Mathematical linguistics." In The Oxford In-
ternational Encyclopedia of Linguistics, 2nd edition, 17-20. Oxford: Oxford University Press, 2003
<http://www.kornai.com/MatLing/matling3.pdf> (accessed December 6, 2011).

Schaner-Wolles, Chris. "Die Sprache des Menschen - auf der Suche nach der Universalgrammatik. Über den kindlichen Spracherwerb und über das Gehirn als zentrales Sprachorgan." In Was ist der Mensch, edited by Karl-Franzens-Universität, 135-170. Graz: Grazer
Universitätsverlag, 2008
<https://fronter.univie.ac.at/links/files.phtml/1521295505$767397201$/
Ressourcen/Literatur+zur+Lehrveranstaltung/Schaner-Wolles+2008+Die+Sprache+des+Menschen.pdf> (accessed December 6, 2011).

Thomas Graf <http://tgraf.bol.ucla.edu/> (accessed December 6, 2011).

Bridges Issues

Podcast