Structural Zeros versus Sampling Zeros
نویسندگان
چکیده
Probabilistic sequence models estimated from large corpora typically require smoothing techniques to reserve some probability mass for unobserved events. These techniques fail to distinguish between events unobserved due to sampling limitations, sampling zeros, and those unobserved due to structural reasons such as syntactic constraints, structural zeros. We investigate the use of statistical tests to determine structural zeros, to avoid assigning them probability mass and thereby improve model accuracy. Experimental results on a context-free parsing task demonstrate the usefulness of these techniques.
منابع مشابه
Bayesian Multiple Imputation for Large-Scale Categorical Data with Structural Zeros
We propose an approach for multiple imputation of items missing at random in large-scale surveys with exclusively categorical variables that have structural zeros. Our approach is to use mixtures of multinomial distributions as imputation engines, accounting for structural zeros by conceiving of the observed data as a truncated sample from a hypothetical population without structural zeros. Thi...
متن کاملImproving the stability properties of sampling zeros of multivariable discrete model via Taylor method and multirate fast sampling
It is well known that multirate input and hold, such as a generalized sample hold function (GSHF), can be used to shift the sampling zeros of a discrete-time model for a continuous-time system. This paper deals with the stability of sampling zeros, as the sampling period tends to zero, of discrete-time models that are composed of a GSHF, a continuous-time multivariable plant and a sampler in ca...
متن کاملFurther Observations on Blocking Zeros in Linear Muitivariabie systems (RESEARCH NOTE).
While attempting to clarify the confusion concerning the conceptualization of "blocking zeros" in state space in the recent literature, some new observations are made on the relationship between pole-zero cancellation and transmission blocking. An important distinction between uncontrollable and unobservable eigenvalue s is pointed out; and it is argued that the description of a Blocking Zero, ...
متن کاملModels for count data with many zeros
Poisson regression models provide a standard framework for the analysis of count data. In practice, however, count data are often overdispersed relative to the Poisson distribution. One frequent manifestation of overdispersion is that the incidence of zero counts is greater than expected for the Poisson distribution and this is of interest because zero counts frequently have special status. For...
متن کاملLINEAR ESTIMATE OF THE NUMBER OF ZEROS OF ABELIAN INTEGRALS FOR A KIND OF QUINTIC HAMILTONIANS
We consider the number of zeros of the integral $I(h) = oint_{Gamma_h} omega$ of real polynomial form $omega$ of degree not greater than $n$ over a family of vanishing cycles on curves $Gamma_h:$ $y^2+3x^2-x^6=h$, where the integral is considered as a function of the parameter $h$. We prove that the number of zeros of $I(h)$, for $0 < h < 2$, is bounded above by $2[frac{n-1}{2}]+1$.
متن کامل