In spite of difficulty in defining the syllable unequivocally, and controversy over its role in theories of spoken and written language processing, the syllable is a potentially useful unit in several practical tasks which arise in computational linguistics and speech technology. For instance, syllable structure might embody valuable information for building word models in automatic speech recognition, and concatenative speech synthesis might use syllables or demisyllables as basic units. In this paper, we first present an algorithm for determining syllable boundaries in the orthographic form of unknown words that works by analogical reasoning from a database or corpus of known syllabifications. We call this syllabification by analogy (SbA). It is similarly motivated to our existing pronunciation by analogy (PbA) which predicts pronunciations for unknown words (specified by their spellings) by inference from a dictionary of known word spellings and corresponding pronunciations. We show that including perfect (according to the corpus) syllable boundary information in the orthographic input can dramatically improve the performance of pronunciation by analogy of English words, but such information would not be available to a practical system. So we next investigate combining automatically-inferred syllabification and pronunciation in two different ways: the series model in which syllabification is followed sequentially by pronunciation generation; and the parallel model in which syllabification and pronunciation are simultaneously inferred. Unfortunately, neither improves performance over PbA without syllabification. Possible reasons for this failure are explored via an analysis of syllabification and pronunciation errors.

This paper describes in detail an algorithm for the unsupervised learning of natural language morphology, with emphasis on challenges that are encountered in languages typologically similar to European languages. It utilizes the Minimum Description Length analysis described in Goldsmith (2001), and has been implemented in software that is available for downloading and testing.

For a random permutation of $n$ objects, as $n \to \infty$, the process giving the proportion of elements in the longest cycle, the second-longest cycle, and so on, converges in distribution to the Poisson–Dirichlet process with parameter 1. This was proved in 1977 by Kingman and by Vershik and Schmidt. For soft reasons, this is equivalent to the statement that the random permutations and the Poisson–Dirichlet process can be coupled so that zero is the limit of the expected $\ell_1$ distance between the process of cycle length proportions and the Poisson–Dirichlet process. We investigate how rapid this metric convergence can be, and in doing so, give two new proofs of the distributional convergence.

One of the couplings we consider has an analogue for the prime factorizations of a uniformly distributed random integer, and these couplings rely on the ‘scale-invariant spacing lemma’ for the scale-invariant Poisson processes, proved in this paper.

We put the final piece into a puzzle first introduced by Bollobás, Erdõs and Szemerédi in 1975. For arbitrary positive integers $n$ and $r$ we determine the largest integer $\Delta=\Delta (r,n)$, for which any $r$-partite graph with partite sets of size $n$ and of maximum degree less than $\Delta$ has an independent transversal. This value was known for all even $r$. Here we determine the value for odd $r$ and find that $\Delta(r,n)=\Delta(r-1,n)$. Informally this means that the addition of an oddth partite set does not make it any harder to guarantee an independent transversal.

In the proof we establish structural theorems which could be of independent interest. They work for all$r\geq 7$, and specify the structure of slightly sub-optimal graphs for even$r\geq 8$.

We show that a maximum cut of a random graph below the giant-component threshold can be found in linear space and linear expected time by a simple algorithm. In fact, the algorithm solves a more general class of problems, namely binary 2-variable constraint satisfaction problems. In addition to Max Cut, such Max 2-CSPs encompass Max Dicut, Max 2-Lin, Max 2-Sat, Max-Ones-2-Sat, maximum independent set, and minimum vertex cover. We show that if a Max 2-CSP instance has an ‘underlying’ graph which is a random graph $G \in \mathcal{G}(n,c/n)$, then the instance is solved in linear expected time if $c \leq 1$. Moreover, for arbitrary values (or functions) $c>1$ an instance is solved in expected time $n \exp(O(1+(c-1)^3 n))$; in the ‘scaling window’ $c=1+\lambda n^{-1/3}$ with $\lambda$ fixed, this expected time remains linear.

Our method is to show, first, that if a Max 2-CSP has a connected underlying graph with $n$ vertices and $m$ edges, then $O(n 2^{(m-n)/2})$ is a deterministic upper bound on the solution time. Then, analysing the tails of the distribution of this quantity for a component of a random graph yields our result. Towards this end we derive some useful properties of binomial distributions and simple random walks.

We study relational structures (especially graphs and posets) which satisfy the analogue of homogeneity but for homomorphisms rather than isomorphisms. The picture is rather different. Our main results are partial characterizations of countable graphs and posets with this property; an analogue of Fraïssé's theorem; and representations of monoids as endomorphism monoids of such structures.

In this paper, we give sharp upper bounds on the maximum number of edges in very unbalanced bipartite graphs not containing any cycle of length 6. To prove this, we estimate roughly the sum of the sizes of the hyperedges in triangle-free multi-hypergraphs.

The adaption of combinatorial duality to infinite graphs has been hampered by the fact that while cuts (or cocycles) can be infinite, cycles are finite. We show that these obstructions fall away when duality is reinterpreted on the basis of a ‘singular’ approach to graph homology, whose cycles are defined topologically in a space formed by the graph together with its ends and can be infinite. Our approach enables us to complete Thomassen's results about ‘finitary’ duality for infinite graphs to full duality, including his extensions of Whitney's theorem.