Informatica

Machine Translation has become an important tool in overcoming the language barrier. The quality of translations depends on the languages and used methods. The research presented in this paper is based on well-known standard methods for Statistical Machine Translation that are advanced by a newly proposed approach for optimizing the weights of translation system components. Better weights of system components improve the translation quality. In most cases, machine translation systems translate to/from English and, in our research, English is paired with a Slavic language, Slovenian. In our experiment, we built two Statistical Machine Translation systems for the Slovenian-English language pair of the Acquis Communautaire corpus. Both systems were optimized using self-adaptive Differential Evolution and compared to the other related optimization methods. The results show improvement in the translation quality, and are comparable to the other related methods.

The problem of recursive estimation of a state of dynamic systems in the presence of time-varying outliers in observations to be processed has been considered. A learning phase used in the state estimation is investigated, assuming that the observations of a noisy output signal and that of a training one are given. A technique based on robust filtering by means of a bank of parallel Kalman filters and on the procedure of optimization of the state estimation itself is used, choosing, at each time moment, a current estimate, that ensures a minimal absolute deviation from the current value of the teaching signal. An approach, based on the relation between the mean squared deviation of state estimates from the true state and innovation sequence variance as well as on the fact that both variables achieve their minimum for the same filter from the respective Kalman filter bank, is proposed here for a working phase, where a training signal will be absent. The recursive technique based on an adaptive state estimation with optimization procedure is worked out. The results of numerical simulation of the linear discrete-time invariant (LTI) system (56) by computer using a bank, consisting of Kalman filters are given (Figs. 1–5).

In the previous paper (Pupeikis, 1998), the problem of recursive estimation of the state of linear dynamic systems, described by an autoregressive model (AR), in the presence of time-varying outliers in observations to be processed has been considered. An approach to the robust recursive state estimation has been obtained and proved by estimating the real chemical process (Box and Jenkins, 1970). The aim of the given paper is the development of the abovementioned approach for the robust recursive state estimation of an autoregressive-moving average (ARMA) process in a case of additive noises with patchy outliers. The results of numerical simulation and the state estimation of the AR model (Figs. 1–4) and the real chemical process, described by the ARMA model, which is chosen from the same book of Box and Jenkins (Figs. 5–8) are given.

In the previous papers (Novovičova, 1987; Pupeikis 1991) the problem of recursive least square (RLS) estimation of dynamic systems parameters in the presence of outliers in observations has been considered, when the filter, generating an additive noise, has a transfer function of a particular form, see Fig. 1, 2. The aim of the given paper is the development of well-known classical techniques for robust on-line estimation of unknown parameters of linear dynamic systems in the case of additive noises with different transfer functions. In this connection various ordinary recursive procedures, see Fig. 2–6, are worked out when systems' output is corrupted by the correlated noise containing outliers. The results of numerical simulation by IBM PC/AT (Table 1) are given.

In the previous paper (Pupeikis, 1992) the problem of off-line estimation of dynamic systems parameters in the presence of outliers in observations have been considered, when the filter generating an additive noise has a very special form. The aim of the given paper is the development, in such a case, of classical generalized least squares method (GLSM) algorithms for off-line estimation of unknown parameters of dynamic systems. Two approaches using batch processing of the stored data are worked out. The first approach is based on the application of S-, H-, W- algorithms used for calculation of M-estimates, and the second one rests on the replacement of the corresponding values of the sample covariance and cross-covariance functions by their robust analogues in respective matrices of GLSM and on a further application of the least squares (LS) parameter estimation algorithms. The results of numerical simulation by IBM PC/AT (Table 1) are given.