Named entity recognition (NER) refers to the identification of proper nouns from natural language text and classifying them into named entity types, such as person, location, and organization. Due to the widespread applications of NER, numerous NER techniques and benchmark datasets have been developed for both Western and Asian languages. Even though Shahmukhi script of the Punjabi language has been used by nearly three fourths of the Punjabi speakers worldwide, Gurmukhi has been the main focus of research activities. Specifically, a benchmark NER corpus for Shahmukhi is non-existent, which has thwarted the commencement of NER research for the Shahmukhi script. To this end, this article presents the development and specifications of the first-ever NER corpus for Shahmukhi. The newly developed corpus is composed of 318,275 tokens and 16,300 named entities, including 11,147 persons, 3,140 locations, and 2,013 organizations. To establish the strength of our corpus, we have compared the specifications of our corpus with its Gurmukhi counterparts. Furthermore, we have demonstrated the usability of our corpus using five supervised learning techniques, including two state-of-the-art deep learning techniques. The results are compared, and valuable insights about the behaviors of the most effective technique are discussed.

Named Entity Recognition (NER) plays a pivotal role in various natural language processing tasks, such as machine translation and automatic question-answering systems. Recognizing the importance of NER, a plethora of NER techniques for Western and Asian languages have been developed. However, despite having over 490 million Urdu language speakers worldwide, NER resources for Urdu are either non-existent or inadequate. To fill this gap, this article makes four key contributions. First, we have developed the largest Urdu NER corpus, which contains 926,776 tokens and 99,718 carefully annotated NEs. The developed corpus has at least doubled the number of manually tagged NEs as compared to any of the existing Urdu NER corpora. Second, we have generated six new word embeddings using three different techniques, fastText, Word2vec, and Glove, on two corpora of Urdu text. These are the only publicly available embeddings for the Urdu language, besides the recently released Urdu word embeddings by Facebook. Third, we have pioneered in the application of deep learning techniques, NN and RNN, for Urdu named entity recognition. Finally, we have performed 10-folds of 32 different experiments using the combinations of a traditional supervised learning and deep learning techniques, seven types of word embeddings, and two different Urdu NER datasets. Based on the analysis of the results, several valuable insights are provided about the effectiveness of deep learning techniques, the impact of word embeddings, and variations of datasets.

Abstract Several organizations maintain textual process descriptions alongside graphical process descriptions to make them usable for all stakeholders. Maintaining textual process descriptions in the presence of continuously changing processes is a labor‐intensive task. Therefore, the automatic generation of textual descriptions is desirable. However, the trade‐offs between the manual and automatic generation of descriptions are yet to be investigated. To that end, this paper aims to answer two vital questions. How similar are the descriptions generated by the two approaches? What is the impact of using the two types of descriptions on process matching? To answer these specific questions, we have generated textual descriptions of 552 process models using the two approaches. To answer the first question, we have applied six text‐matching techniques and established that the descriptions overlap significantly; however, the formulation of sentences is substantially different. For answering the second question, we have used 11 text‐matching techniques to evaluate the impact of both descriptions on process matching. Results show (a) the choice of matching technique, and the type of description, have an impact on the matching performance and (b) vector space model (VSM) is the most appropriate matching technique whereas 5 gram is the worst performing technique.

The ability of tiny embedded devices to run large feature-rich programs is typically constrained by the amount of memory installed on such devices. Furthermore, the useful operation of these devices in wireless sensor applications is limited by their battery life. This paper presents a call stack redesign targeted at an efficient use of RAM storage and CPU cycles by a Java program running on a wireless sensor mote. Without compromising the application programs, our call stack redesign saves 30% of RAM, on average, evaluated over a large number of benchmarks. On the same set of bench-marks, our design also avoids frequent RAM allocations and deallocations, resulting in average 80% fewer memory operations and 23% faster program execution. These may be critical improvements for tiny embedded devices that are equipped with small amount of RAM and limited battery life. However, our call stack redesign is equally effective for any complex multi-threaded object oriented program developed for desktop computers. We describe the redesign, measure its performance and report the resulting savings in RAM and execution time for a wide variety of programs.

The ability of tiny embedded devices to run large and feature-rich Java programs is typically constrained by the amount of memory installed on those devices. Furthermore, the useful operation of such devices in a wireless sensor application is limited by their battery life. We propose a garbage collection (GC) scheme called Offline GC which alleviates both these limitations. Our approach defies the current practice in which an object may be deallocated only if it is unreachable. Offline GC allows freeing an object that is still reachable but is guaranteed not to be used again in the program. Furthermore, it may deallocate an object inside a function, a loop or a block where it is last used, even if that object is assigned to a global field. This leads to a larger amount of memory available to a program.

We present TakaTuka, a tiny Java Virtual Machine (JVM) for wireless sensor motes. TakaTuka's preliminary version successfully runs on Crossbow's mica2 motes. Furthermore, TakaTuka also runs on Windows and Unix.

For label switched networks, such as MPLS and GMPLS, most existing traffic engineering (TE) solutions work in a single routing domain. These solutions do not work when a route from the ingress node to the egress node leaves the routing area or the autonomous system (AS) of the ingress node. In such cases, the path computation problem becomes complicated because of the unavailability of the complete routing information throughout the network. We present CWS (computation while switching), a new inter-domain path computation scheme which tries to compute a near-optimal path without assuming the availability of complete topology information. We provide a detailed comparison of the CWS scheme with another per-domain path computation scheme given in J.-P. Vasseur et al. (2006). Unlike the standard per-domain path computation scheme (Vasseur et al., 2006), the CWS scheme continues the quest for a better path instead of terminating the search at the first available path, resulting a significant improvement in terms of path optimality. In particular, CWS guarantees that, for a given network state, a computed inter-domain path will traverse a minimum number of domains. This improvement in path computation directly impacts the amount of traffic that can be allowed on the network. For example, for the COST266 topology with 28 domains and 37 bidirectional inter-domain links, CWS places 960 of the requested 2000 paths as compared to 683 paths placed by existing schemes. Finally, the path setup latency of the CWS scheme remains comparable to that of existing schemes, by allowing the data flow as soon as the first feasible path is found.

In label-switched networks such as MPLS, protection routing involves computing and setting up the backup paths at the same time when the primary paths are routed. It has previously been shown that two or more backup paths may share bandwidth along common links if such backup paths will never be activated simultaneously. Such sharing between the backup paths leads to reduced bandwidth reservations and, hence improved performance in terms of number of path requests that can be accommodated on the network (Aslam et al., 2005). We present a novel idea that backup paths may also share bandwidth with certain primary paths, thereby further reducing the overall bandwidth reservations on the network. This results in even more path requests being accommodated on the network. Sharing with primary paths is possible with any protection routing framework. To demonstrate this sharing, we use the NPP protection routing framework as an example (Aslam et al., 2005). We provide the enhancements to the NPP framework needed to exploit sharing with the primary paths. For the enhanced NPP framework, simulation results on various networks confirm that sharing with primary paths indeed results in better network utilization. This increased performance is achieved with bounded local state information and without requiring any additional routing or signaling overhead.

We investigate the problem of distributed online routing of bandwidth guaranteed paths with local restoration. A unified model is proposed that captures the bandwidth sharing characteristic of backup paths that provision local restoration, corresponding to different fault models. We apply the model to describe bandwidth sharing on backup paths for varying degrees of network state information. The extent of backup bandwidth sharing depends on the amount of network state information made available through routing protocols. A key design criterion for traffic engineering schemes is to maximize the sharing between backup paths, while minimizing this protocol overhead. M.S. Kodialam and T.V. Lakshman (see Proc. Infocom, p.376-85, 2001) demonstrated that propagating a constant amount of aggregated information per link leads to cost effective bandwidth sharing. We propose oAIS, a new aggregate information scenario, in which we judiciously select the propagated information, such that the protocol overhead is identical to that of Kodialam and Lakshman. Simulations show that oAIS outperforms other information scenarios with comparable protocol overheads.