It is unlikely that an assiduous researcher has many moments in his career when a problem he has long been wrestling with is suddenly resolved. I had one such moment on March 6, 1989. But first some background.

In the Finnish sentence “Kaksi alusta upposi,” the word “alusta” can be interpreted in several ways. It can be one of several forms of the verb “alustaa - format,” the elative case of the noun “alku - beginning,” or the partitive of the nouns “alunen - coaster” and “alus - craft.” How should one reach the correct inference? Another issue is that every word form has some syntactic function; depending on the word and its form there can be over ten of them. A noun can be, to name a few possibilities, a subject (Leena tuli - Leena came), object (Tapasin Leenan - I met Leena), a premodifier (Leenan kirja - Leena’s book) and a complement to a postposition in Finnish (Leenan kanssa) or preposition in English (with Leena). How should one decide which is the correct function in each instance?

For years I applied various solutions to these two problems by combining my knowledge of the theory of syntax with my steadily growing programming skills. I attempted to draw up rules that would immediately deduce the correct morphological interpretation and grammatical function in each case. I did not make the kind of progress I had hoped for.

March 6, 1989 at 9.30 AM: EUREKA! I was on my way to a meeting in Dublin, where the SIMPR-project (Structured Information Management, Processing, and Retrieval) was about to start as part of the European Union’s ESPRIT II. About fifteen parties were involved in the project, two of them from Finland: the Nokia Corporation and the Department of General Linguistics at the University of Helsinki, particularly the Centre of Excellence on computational linguistics. It was the Finnish version of David and Goliath!

On that day in March, I first flew to Amsterdam and then took another flight to Dublin. When the plane to Dublin was over the North Sea, I suddenly came up with a single general solution to my two morphosyntactic problems. You had to do exactly the opposite of what I (along with everyone else) had been trying to do. You had to remove the wrong options instead of looking for the correct one right away.

In half an hour I wrote down the basic principles of Constraint Grammar on a piece of cross-ruled paper. When I arrived back home, I started febrile programming, and a long series of tests began. Vital assistance and additional ideas were provided by Arto Anttila (who is currently professor of linguistics at Stanford University), Juha Heikkilä and Atro Voutilainen. I published the results at the International Conference on Computational Linguistics in 1990. This six-page publication is the most cited scientific article I have produced: “Constraint Grammar as a framework for parsing running text.”

Later on, Constraint Grammar has been implemented several times, and it has been applied to over ten languages. The model is also used in several commercial applications.

My other eureka moment is also related to syntax. In 2007, after working for nearly ten years, I managed to empirically prove that the syntax of natural languages does not recognise the construct of the “unbounded centre-embedding of clauses.” That is why the formal language type of syntax is finite state: “Constraints on multiple centre-embedding of clauses.”