Higher-order Lexical Semantic Models for Non-factoid Answer Reranking

Lexical semantic models provide robust performance for question answering, but, in general, can only capitalize on direct evidence seen during training. For example, monolingual alignment models acquire term alignment probabilities from semi-structured data such as question-answer pairs; neural network language models learn term embeddings from unstructured text. All this knowledge is then used to estimate the semantic similarity between question and answer candidates. We introduce a higher-order formalism that allows all these lexical semantic models to chain direct evidence to construct indirect associations between question and answer texts, by casting the task as the traversal of graphs that encode direct term associations. Using a corpus of 10,000 questions from Yahoo! Answers, we experimentally demonstrate that higher-order methods are broadly applicable to alignment and language models, across both word and syntactic representations. We show that an important criterion for success is controlling for the semantic drift that accumulates during graph traversal. All in all, the proposed higher-order approach improves five out of the six lexical semantic models investigated, with relative gains of up to +13% over their first-order variants.

Analytical formula for electron-impact ionization cross section in the one Coulomb wave model

In the framework of the one Coulomb wave model, an analytical formula is derived for the triple differential cross section for the ionization of atomic target by electron impact. Both the direct and exchange terms of the transition amplitude are considered. The exchange term is calculated in the present work. However, the direct term is deduced from our previous work (Sahlaoui and Bouamoud. Can. J. Phys. 88, 905 (2010)). The obtained results can be used to compute the transition amplitude defined in others models, as for example, the distorted wave Born approximation and the Coulomb two-wave model. The mathematical techniques developed here can be also used to compute other complicated integrals that can be encountered in atomic physics and quantum chemistry.

Free-Form Modeling in Bilateral Brep and CSG Representation Schemes

This paper presents a unified approach for incorporating free-form solids in bilateral Brep and CSG representation schemes, by resorting to low-degree (quadratic, cubic) algebraic surface patches. We develop a general CSG solution that represents a free-form solid as a boolean combination of a direct term and a complicated delta term. This solution gives rise to the trunctet-subshell conditions, under which the delta term computation can be obviated. We use polyhedral smoothing to construct a Brep consisting of quadratic algebraic patches that meet with tangent-plane continuity, such that the trunctet-subshell conditions are guaranteed automatically. This guarantee is not currently available for cubic patches. The general CSG solution thus applies whenever trunctet-subshell conditions are violated, e.g. sometimes for cubic patches or sometimes for patches of any degree that are subject to shape control operations. Manifold solids of arbitrary topology can be represented in our dual representation system. Ensuing CSG constructs are parallel processed on the RayCasting Engine to support a wide range of solid modeling applications, including general sweeping, Minkowski operations, NC machining, and touch-sense probing.

Mechanics of stretch in activated crustacean slow muscle. II. Dynamic changes in force in response to stretch

1. The mechanical dynamics of the ventral superficial muscles (VSM) of the abdomen of the hermit crab, Pagurus pollicarus, have been analyzed to develop a quantitative model of gradedly excitable arthropod muscle. Such a model is important for understanding the role of proprioceptive reflexes in posture and movement. 2. The decay in force produced after ramp stretch of both passive and active muscle was approximated by the use of regression equations involving a direct term and one to three exponential terms. A second-order equation produced an acceptable description of this decay over short (0.5 s) sampling durations. 3. The rate constants of the regression equation did not vary with stretch length, velocity, or activation level of the muscle. For the two-exponential-term model, the rate constants were approximately 90 and 9 s-1 for a sample duration of 0.3 s. An additional rate constant of approximately 1 s-1 was needed to adapt the model to longer sample times. 4. The direct term and the middle-order (9 s-1) residual were both functions of stretch length and activation level. The high-order (90 s-1) residual was primarily a function of stretch length and velocity. Transfer functions omitting the velocity dependence adequately described the mechanical dynamics of the muscle for physiological ranges of stretch velocity. 5. White-noise length perturbations were used to calculate spectral density functions of muscle force and length. These measurements confirmed the principal observations of the ramp stretch analysis: the frequency response of the muscle was independent of the level of activation; the magnitude of the stiffness increased over the stretch frequency range of 4-40 Hz and was then almost constant; and the phase response of the muscle became slightly positive over the same range of stretch frequency. 6. The speed of activation of the muscle to different stimulus frequencies was estimated by fitting a single exponential equation to the rise in isometric tension at the onset of stimulation of the motor nerve. The rate constant increased with stimulus frequency, but its maximum value was only 1.8 s-1, about one-fourth of the middle mechanical rate constant. 7. Because muscle activation is slower than the mechanical dynamics, it is unlikely that the nervous system can regulate muscle dynamics. However, it is possible that mechanical impedance could be regulated to maintain a desired time-averaged value.

Band Structure of CaCd and CaTl and the Knight Shift of 113Cd- and 205Tl-NMR in the System CaCd1-x Tlx

The energy bands of ordered CaCd and CaTl have been calculated by the nonrelativistic augmented plane wave (APW) method. The electron structure in the system CaCd1-xTlx is deduced from these calculations by using the rigid band model for the phases with 0<x<1. The band structures of CaCd and CaTl are similar to the valence bands of other phases of the CsCl-type. From the energy eigenvalues the electronic density of states curve, the partial densities of states curves, and the Fermi energy have been obtained. For states near the Fermi surface the spin density at the position of the Cd- and Tl-nuclei has been determined. The Knight shift Ks of the 113Cd-NMR and the 205Tl-NMR in the system CaCd1-xTlx has been calculated as a function of x. The slope of the curve Ks (x) for the Cd-NMR is equal for experimental and theoretical results. The absolute value of the calculated Knight shift is about a factor of 1.4 too small. Only the direct term to the Knight shift has been calculated. Relativistic effects have been included by a scale factor. It has not been possible to explain the shape of the function (2) for the Tl-NMR, since a full relativistic APW calculation is necessary for CaTl