English to Luganda SMT: Ganda Noun Class Prefix Segmentation for Enriched Machine Translation

Abstract

Luganda or Ganda is a morphologically rich and low-resource language from Uganda. The morphological richness of Luganda sentences has an impact on the quality of translation and this work looks at improving machine translation (MT) for English to Luganda. Luganda sentence formation bases on 10 noun classes with a prefix for singular and plural. In various aspects, the interaction of these class prefixes in sentences usually enforces different words such as nouns, verbs, adverbs and adjectives to be in agreement with the subject in a given sentence. Such scenarios have resulted in various Luganda word inflectional and derivational tendencies because each noun, verb, adverb or adjective finds itself having to combine with a prefix from a class it belongs and thus creating new word forms. Using 6 statistical machine translation (SMT) models divided equally into base and morphology models, we propose a procedure to segment Luganda sentences from our English-Luganda parallel Bible corpus. Our morphological segmentation approach bases on Ganda Noun Class (GNC) prefixes and we design a tool we call GandaKIT to segment Luganda sentences at pre-processing stage and desegement them after translation at post processing stage. In experiments, we compare translation performance of SMT base models against systems trained with morphological segmentation at pre-processing stage. Our results show an improvement in MT performance over base models ranging by a difference of 1.58 BLEU points and 0.2257 NIST score for our best system.

Monitoring and Forecasting Model State of the Telecommunications Network using Fuzzy Neural Networks

“…It is calculated using a modified reverse propagation method, which is called the reverse propagation method with network deployment in time [11]. As in the reverse propagation method for direct propagation networks, the process of calculating the gradient (weight changes) occurs in three stages: Direct pass -calculating the state of layers; Reverse pass -calculating the error of layers; Calculation of weight changes based on data obtained at the first and second stages [31]. Monitoring and forecasting the state of a telecommunications network involves studying a whole group of issues related to the network infrastructure (routers, switches), traffic structure, network load, configuration, security, quality of service, and so on.…”

Proposing an Elliptic Curve Cryptosystem with The Symmetric Key for Vietnamese Text Encryption and Decryption

“…However, when we are working over a finite field, for example Fp, this is not a problem because we can continually reduce mod p and keep the numbers involved relatively small. It should be noted that the associative law allows us to make these computations without worrying about what order we use to combine the summands [11,12]. When the sum of the points P and Q on the elliptic curve E is shown in Figure 1.…”