Researcher profile

Stephen Cox

Stephen Cox contributes to research discovery and scholarly infrastructure.

ResearcherAffiliation not importedOpen to collaborate

Computation and Language math.AP

Trust snapshot

Quick read

Trust 13 - Baseline

2works

0followers

2topics

4close collaborators

Actions

Decide how to stay connected

Follow researcher0

Open graph Browse works

Inspect adjacent work, topics, institutions and collaborators without jumping out to a separate graph page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2016arXiv

Bi-Text Alignment of Movie Subtitles for Spoken English-Arabic Statistical Machine Translation

We describe efforts towards getting better resources for English-Arabic machine translation of spoken text. In particular, we look at movie subtitles as a unique, rich resource, as subtitles in one language often get translated into other languages. Movie subtitles are not new as a resource and have been explored in previous research; however, here we create a much larger bi-text (the biggest to date), and we further generate better quality alignment for it. Given the subtitles for the same movie in different languages, a key problem is how to align them at the fragment level. Typically, this is done using length-based alignment, but for movie subtitles, there is also time information. Here we exploit this information to develop an original algorithm that outperforms the current best subtitle alignment tool, subalign. The evaluation results show that adding our bi-text to the IWSLT training bi-text yields an improvement of over two BLEU points absolute.

preprint2014arXiv

Orientation-dependent pinning and homoclinic snaking on a planar lattice

We study homoclinic snaking of one-dimensional, localised states on two-dimensional, bistable lattices via the method of exponential asymptotics. Within a narrow region of parameter space, fronts connecting the two stable states are pinned to the underlying lattice. Localised solutions are formed by matching two such stationary fronts back-to-back; depending on the orientation relative to the lattice, the solution branch may `snake' back and forth within the pinning region via successive saddle-node bifurcations. Standard continuum approximations in the weakly nonlinear limit (equivalently, the limit of small mesh size) do not exhibit this behaviour, due to the resultant leading-order reaction-diffusion equation lacking a periodic spatial structure. By including exponentially small effects hidden beyond all algebraic orders in the asymptotic expansion, we find that exponentially small but exponentially growing terms are switched on via error function smoothing near Stokes lines. Eliminating these otherwise unbounded beyond-all-orders terms selects the origin (modulo the mesh size) of the front, and matching two fronts together yields a set of equations describing the snaking bifurcation diagram. This is possible only within an exponentially small region of parameter space---the pinning region. Moreover, by considering fronts orientated at an arbitrary angle $ψ$ to the $x$-axis, we show that the width of the pinning region is non-zero only if $\tanψ$ is rational or infinite. This is the first time a formula explicitly relating the orientation of a front to the width of its pinning region has been derived. The asymptotic results are compared with numerical calculations, with good agreement.