Is Recurrence Redundant? Revisiting Allen and Christiansen (1996)
نویسنده
چکیده
Several influential connectionist models of the word segmentation task (e.g. Cairns, Shillcock, Chater, & Levy, 1997; Christiansen, Allen, & Seidenberg, 1998) follow Elman (1990) in using simple recurrent networks (SRNs). The use of SRNs in this context appears to be traditional rather than independently motivated. This paper investigates whether alternatives to SRNs can achieve similar performance. Specifically, it reports a replication of part of Allen and Christiansen (1996), and a comparison with multi-layer perceptrons (MLPs). The major results are confirmed; however, the SRN topology is shown not to be necessary for the task addressed. MLPs perform equally well, even when they have access to less context.
منابع مشابه
MTBF evaluation for 2-out-of-3 redundant repairable systems with common cause and cascade failures considering fuzzy rates for failures and repair: a case study of a centrifugal water pumping system
In many cases, redundant systems are beset by both independent and dependent failures. Ignoring dependent variables in MTBF evaluation of redundant systems hastens the occurrence of failure, causing it to take place before the expected time, hence decreasing safety and creating irreversible damages. Common cause failure (CCF) and cascading failure are two varieties of dependent failures, both l...
متن کاملRevisiting Beta 2 Glycoprotein I, the Major Autoantigen in the Antiphospholipid Syndrome
Beta 2 glycoprotein I (β2GPI) is a single chain 50 kDa highly glycosylated glycoprotein at an approximate concentration of 4 μM in cells. The abundance of this protein in plasma and its high state of preservation indicate the important role of this protein in mammalian. In addition, β2GPI has a particular structure in the fifth domain, and is categorized as the major antigen recognized by autoa...
متن کاملOn the optimality of Allen and Kennedy's algorithm for parallelism extraction in nested loops
We explore the link between dependence abstractions and maximal parallelism extraction in nested loops. Our goal is to find, for each dependence abstraction, the minima] transformations needed for maximal parallelism extraction. The result of this paper is that Allen and Kennedy's algorithm is optimal when dependences are approximated by dependence levels. This means that even the most sophisti...
متن کامل