From: Subject: The Heinz Electronic Library Interactive Online System (HELIOS): Building a Digital Archive Using Imaging, OCR, and Natural Language Processing Technologies Date: Fri, 24 Jul 2009 13:02:10 -0400 MIME-Version: 1.0 Content-Type: multipart/related; type="text/html"; boundary="----=_NextPart_000_000B_01CA0C5E.F4E57CE0" X-MimeOLE: Produced By Microsoft MimeOLE V6.00.2900.5579 This is a multi-part message in MIME format. ------=_NextPart_000_000B_01CA0C5E.F4E57CE0 Content-Type: text/html; charset="utf-8" Content-Transfer-Encoding: quoted-printable Content-Location: http://epress.lib.uh.edu/pr/v6/n4/gall6n4.html =EF=BB=BF The Heinz Electronic Library Interactive Online = System (HELIOS): Building a Digital Archive Using Imaging, OCR, and = Natural Language Processing Technologies =20

Galloway, Edward A., and Gabrielle V. Michalek. "The Heinz = Electronic=20 Library Interactive Online System (HELIOS): Building a Digital Archive = Using=20 Imaging, OCR, and Natural Language Processing Technologies." The=20 Public-Access Computer Systems Review 6, no. 4 (1995).

1.0 Introduction

In February 1994, Carnegie Mellon University (CMU) embarked on an = ambitious=20 project to convert one million pages of the congressional papers of = Senator John=20 Heinz (R-PA) into digital format and to provide access to these papers = through=20 innovative information retrieval software developed at CMU. Named in = memory of=20 the late Senator, the Heinz Electronic Library Interactive Online System = (HELIOS) supports full-page digital images and it utilizes natural = language=20 processing (NLP) technology to search large quantities of unstructured = text.=20 HELIOS will allow researchers to access the Heinz papers through the = campus=20 network as well as through the Internet.=20

Over one million dollars was donated by the Heinz Family Foundation, = Heinz=20 Company Foundation, and Heinz Endowments to support the establishment of = the H.=20 John Heinz III Archives and the digitization project. Heinz assistance = has made=20 it possible to advance the principles of digital preservation and access = for=20 archival collections. In addition to the Heinz gift, CMU has committed = an=20 additional $450,000 in matching resources to the project. These = resources=20 primarily come in the form of permanent full-time staff salaries, = archival=20 equipment, and rental of a processing facility.=20

Our goal is to develop a digital archive that will serve as a model = for the=20 archival profession. We expect to create an archival information = technology=20 environment that dramatically increases the depth of indexing and the = quality of=20 retrieval beyond what archiving resources have traditionally allowed.=20

To create the HELIOS database, documents are scanned, converted to = ASCII form=20 via OCR, verified and organized, and indexed using the CLARIT natural = language=20 processing software. The project will develop three graphical user = interfaces in=20 a Microsoft Windows environment: a scanning interface, an = archivist/verification=20 interface, and an end-user interface.=20

HELIOS represents a significant breakthrough technology that has the=20 potential to transform the work of archivists by helping them to = overcome the=20 significant challenges they face, including an inability to: 1. create = good=20 finding aids and indexes for paper archives that provide deep access to=20 collections, 2. provide effective retrieval from paper archives due to = the=20 inherent diversity and size of these one-of-a-kind files, and 3. offer = broad=20 public access to archives because they represent resources that the = researcher=20 must visit in order to use effectively.=20

Archivists have resisted the use of information technology because = they lack=20 appropriate tools to automatically process large amounts of text for = retrieval.=20 HELIOS will offer such a tool.=20

Clearly, there are many problems yet to be solved in the management = and=20 preservation of digital archives, but it is CMU's intention to work with = the=20 larger archival and library community to help establish standard = practices for=20 digitizing paper archives and to develop the information management = tools to=20 give scholars and students state-of-the-art access to them.=20

2.0 H. John Heinz III Congressional Collection

Shortly after the tragic death of Senator Heinz in 1991, the family = placed=20 the congressional papers at Carnegie Mellon University to serve as the = research=20 centerpiece for the Heinz Graduate School of Public Policy and = Management. CMU=20 spent $70,000 to prepare an archival facility near campus in = anticipation of=20 receiving the collection. Upon completion of the space, the collection = was=20 transferred from its storage facility in Harmarville, Pennsylvania to = CMU.=20

In addition to documenting Heinz' tenure as a three-term member of = the U.S.=20 House of Representatives, the papers focus on his fifteen-year Senate = career.=20 Senator Heinz earned a national reputation based on his work on = retirement and=20 aging concerns, health care, international trade and finance, human = development,=20 and environmental issues. The Heinz papers present a rich and valuable = source of=20 information about the professional life of John Heinz in the U.S. = Congress and=20 the social and political concerns of the nation during the Senator's = tenure. The=20 Heinz Archives will aid scholars in understanding the Senator's = contributions to=20 national policy and allow current public policy makers to build upon his = accomplishments and unfinished work.=20

The H. John Heinz III Archives will provide both traditional and = electronic=20 access to the papers. The Heinz Archives staff uses conventional = processing=20 methods to arrange and describe the papers while applying fundamental=20 preservation techniques to the original material to ensure longevity. = The=20 material is housed in proper environmental conditions suitable for = long-term=20 preservation. To date, over 500 of the original 1,200 cubic feet of = material has=20 been processed. At the completion of processing, the collection should = comprise=20 approximately 650 to 700 cubic feet of material. The HELIOS project will = provide=20 electronic access to the most important series and subseries in the = collection.=20

3.0 HELIOS Team Members

The HELIOS project is comprised of three umbrella units representing = several=20 disciplines. While bringing its own expertise to the creative process, = each unit=20 is making major contributions to the design, creation, and = implementation of=20 HELIOS.=20

3.1 Laboratory for Computational Linguistics

CMU's Laboratory for Computational Linguistics (LCL) focuses its = research=20 efforts on information retrieval issues. LCL researchers developed = efficient=20 methods to analyze and extract language using computers, and this NLP = research=20 is the basis of the CLARIT software.=20

3.2 CLARITECH Corporation

The CLARITECH Corporation, a CMU spin-off company, has improved and = marketed=20 LCL's NLP technology, dubbing it CLARIT. CLARITECH's primary = contribution to the=20 HELIOS project is system design. It is responsible for incorporating = elements of=20 the CLARIT system designed in the LCL into HELIOS, for creating three = graphical=20 user interfaces for the system, and for supporting the development of = new CLARIT=20 tools for HELIOS throughout the duration of the project.=20

3.3 Carnegie Mellon University Libraries

Three different units within the Carnegie Mellon University Libraries = play a=20 major role in the interdisciplinary functioning of the project.=20

The Library Administration is responsible for providing the = leadership=20 function for HELIOS as well as the fiscal management of the project.=20

The Department of Library Information Technology is responsible for=20 developing the HELIOS client/server system, maintaining the system, = training=20 users, and documenting the system.=20

The University Archives facilitates the interdisciplinary teamwork of = the=20 project. The Heinz Archives, a unit within the University Archives, is=20 responsible for establishing control over the collection, appraising and = processing the original Heinz papers, creating a finding aid to the = collection,=20 providing quality reference service, disseminating and cataloging the = collection=20 via OCLC, and preserving the original collection in perpetuity.=20

Working together, the University and Heinz Archives are responsible = for=20 developing three interface specifications, testing the interfaces before = release, scanning the original material into electronic format, = verifying the=20 quality of the images, performing additional organizational tasks, = creating=20 annotations and links to other parts of the electronic collection, = conducting=20 user protocol testing, and training other library staff to use the = system.=20

4.0 HELIOS Document Processing

In the HELIOS project, the original paper documents are processed = based on=20 the typical arrangement scheme of a congressional collection. After this = is=20 completed, document pages from complete series and subseries are = scanned,=20 resulting in 400 dpi image files. The image files are then converted to = ASCII=20 text files using the TextBridge OCR package. The images and text are = verified=20 and additional notes and organization added. The text is indexed by the = CLARIT=20 natural language processing software, resulting in the searchable HELIOS = database.=20

4.1 Scanning

Once an entire series has been processed, the documents are = transferred into=20 electronic format. With the use of a 66 MHz 486 DX2 PC with a 2 GB hard = drive,=20 20" monochrome monitor (1600 x 1200 pixels), and a high-end Fujitsu = scanner, we=20 are creating 400 dpi bitonal TIFF images. The images are compressed = using CCITT=20 Group IV, an international compression standard, and backed-up onto 4 mm = digital=20 data storage tapes.=20

Because the scanning procedure represents the most crucial aspect of = the=20 project, we designed a scanning interface to facilitate the rapid = scanning of=20 documents while capturing essential contextual information for the = end-user.=20

The scanning interface imitates the standard archival collection = arrangement,=20 organizing documents into subgroups, series, subseries, and smaller = units.=20 Operators select the appropriate level using drop-down menus. The = operator then=20 enters the box and folder number as well as the folder title and date. = The=20 scanning interface was also designed to capture "bundles"; that is, = groups of=20 documents originally fastened together by paper clips, staples, or = rubber bands.=20 These bundles, which often reflect inherent meaning, are more difficult = to=20 depict to an online user; however, doing so is important because it = gives the=20 user the same context as if he or she were physically examining the = material.=20

The document feature allows the operator to choose from a prepared = list of=20 document types, such as correspondence, memoranda, speeches, and notes, = and to=20 assign a corresponding date. There are two reasons for doing this. = First,=20 tagging this kind of data will enable a user to restrict a search to a = specified=20 document type. Second, most archival documents do not have distinct = titles. To=20 overcome this problem and generate a useful description of the retrieved = document for the user, the document type and date can be offered as the = title.=20 Providing this kind of "fielded" information is vital for access to the = material=20 and to maintain contextual accuracy.=20

In addition to capturing the contextual information, this interface = was=20 developed to take into account the unique characteristics of archival = documents.=20 Prior to scanning a document, the operator must specify page size; = brightness=20 and contrast quality levels; whether the document is single- or = double-sided;=20 orientation of the page; and the scanner source (flatbed or feeder). = When a page=20 is scanned, it appears in an image viewer, allowing the operator to = determine=20 the success of the scan and to rescan if necessary. Each scanning = session is=20 logged to monitor quality control and record scanning performance.=20

4.2 Optical Character Recognition (OCR)

In order for the system to provide innovative searching capabilities, = the=20 images must be converted to machine-readable format. This text = recognition=20 process, commonly referred to as optical character recognition (OCR), = produces a=20 standard ASCII text file. An off-the-shelf package called TextBridge, a = Xerox=20 Imaging Systems product, is used for OCR conversion. The HELIOS system = designers=20 have tweaked TextBridge to run the OCR process in batch mode at night to = economize staff time and computer usage. They have also implemented a = CLARIT=20 tool to perform post-OCR correction, thereby boosting the accuracy of = the OCR=20 process to an even greater degree.=20

4.3 Verification

Once each page image has a corresponding text file, the archives = staff will=20 utilize an archivist/verification interface that is presently being = designed to=20 support image and text verification, annotation, and organization. To = perform=20 the majority of the verification tasks, operators will use a Sun = SparcStation 20=20 workstation with an 18 GB external hard drive and 20" color monitor = (1280 x 1024=20 pixels). This workstation will also serve as the HELIOS search engine = server and=20 file server. Two Sun SparcStation 5 workstations will be used to perform = additional verification.=20

The archivist/verification interface will display each page image and = its=20 associated ASCII text to the operator, and it will:=20

The Heinz Archives Assistant and graduate students = will=20 perform the majority of the verification tasks. For documents not = scanned in=20 their entirety, such as government publications, operators will note the = availability of the complete report in a regional repository or provide = some=20 other explanation. For poor or skewed images, operators will indicate = that the=20 original page possessed these characteristics.=20

The Heinz Archivist will provide additional organizing, indexing, and = annotations. Descriptive or structural notes will be added to various = levels of=20 the electronic collection. The Heinz Archivist will be able to organize = the=20 electronic archives along several different dimensions apart from its = processed=20 arrangement. For example, a taxonomy of related terms describing any = level of=20 the collection could be constructed. Each series will be linked to its=20 appropriate inventory and description, and cross-reference notes will be = established to link records to similar groups of material.=20

4.4 NLP Search Engine

The HELIOS search engine utilizes natural language processing (NLP)=20 technology. NLP stems from work done in the fields of computer science,=20 artificial intelligence, and linguistics. Natural language is simply = common,=20 everyday language we use to speak and write. Natural language processing = allows=20 users to interact with a computer system, describing topics of interest = using=20 their own language as opposed to reacting to menus and prompts or using = keyword=20 and Boolean searching techniques. Consequently, they can make better use = of the=20 database with only a general knowledge of its contents.=20

As the HELIOS search engine, CLARIT supports more accurate, = sensitive, and=20 robust content-based indexing and retrieval than is possible with = traditional=20 "word-based" information retrieval technologies. Its indexing and = retrieval=20 capabilities are not based on locating individual words, but rather on=20 extracting concepts that accurately characterize the content of = documents.=20 Combined with specialized statistical methods, CLARIT analyzes a query=20 linguistically, comparing it with a similar linguistic analysis of the = actual=20 documents in the database. We have applied CLARIT to the problem of = managing=20 compound documents (text and images) and the special requirements of = archival=20 material.=20

Why use NLP? Concrete disciplines, such as the medical and legal = professions,=20 often communicate and express ideas in rigorous terminology. But = historians and=20 other scholars, who use archives and historical material, approach their = discipline with more imprecise language. This is why NLP technology has = such=20 promise for robust retrieval of archival material.=20

The Text Retrieval Conference (TREC) studies sponsored by the = National=20 Institute of Science and Technology (NIST) and the DOD's Advanced = Research=20 Projects Agency (ARPA) have now demonstrated that CLARIT has a = compelling=20 advantage over traditional keyword and Boolean searching and retrieval. = [1]=20 Studies of keyword and Boolean retrieval systems have shown that = sometimes they=20 provide good precision and sometimes good recall, but never both = together, and=20 often neither. [2] The non-expert searcher (i.e., the average library = user) has=20 even less success. In addition, Boolean logic operators or special = devices like=20 adjacency and nesting are usually ignored by the general user who opts = for=20 single-term searches in hopes of getting the greatest number of = retrieved items.=20 They know from experience that they will do better by manually sifting = the=20 results and selecting relevant documents.=20

Efforts to enhance online records have improved recall at the expense = of=20 precision. Unless we find new tools, moving to full-text electronic = access will=20 only make matters worse. CMU believes that CLARIT is the "better = mousetrap"--one=20 that will be especially useful for accessing archival material.=20

4.5 Prototype End-User Interface

One year ago, approximately 20,000 pages of archival material related = to the=20 work of Michael Lockerby, a Legislative Assistant (1977-1981) = specializing in=20 environmental issues and legislation, were used as a testbed for initial = scanning and CLARIT retrieval.=20

This scanning project allowed us to assess the physical demands of = scanning=20 archival documents and to determine the strengths and weaknesses of=20 off-the-shelf technology used in HELIOS. The resultant document database = was=20 very useful in conducting focus groups on the prototype HELIOS end-user=20 interface, and it was also used for HELIOS demonstrations at other = sites.=20

The prototype end-user interface has four major windows.=20

  1. Query Window. The user enters a natural language search in = this=20 window, such as "the economic impact of environmental regulations and = policies=20 on the steel industry."=20

  2. CLARIT Results Window. This window shows the list of = retrieved=20 pages. The pages are ranked in order of their estimated relevance to = the=20 query. The Change column indicates the upward or downward movement of = pages=20 based on a prior query search (e.g., +12 or -8), and an asterisk = indicates=20 that the system retrieved a new page. The Document Type and Date = fields are=20 used to generate a useful description of the retrieved page, such as=20 "Memoranda--May 5, 1990" or "Speech--November 12, 1990."=20

  3. Document Window. This window provides an ASCII version of = the page=20 with highlighting of retrieval terms associated with the query.=20

  4. Image Viewer. This window displays the bitmapped image of = the=20 original page retrieved for the search.
The end-user will = have the=20 ability to move forward and backward through any level of the collection = (e.g.,=20 move to the next or previous page, the next or previous document, or the = next or=20 previous folder). Since document types will be tagged, the user could = restrict a=20 search to a specified date and document type, such as correspondence, = memoranda,=20 or speeches.=20

CLARIT offers several tools to improve query results. One feature can = extract=20 related terminology from the actual documents. These terms are generated = "on-the-fly" by CLARIT. This feature allows the documents to describe = themselves=20 and eliminates the need for pre-existing indices. A second feature = allows the=20 researcher to use an existing page as an example query to locate more = documents=20 like it. A third tool incorporates a hypertext feature to link other = relevant=20 portions of the collection together.=20

Unless specified, CLARIT searches the entire collection regardless of = the=20 arrangement and contextual framework imposed by the archivist. = Therefore, it is=20 crucial for the user to understand the context in which the retrieved = documents=20 were created. The HELIOS end-user interface will present the name of the = folder=20 from which a retrieved page originated, and it will allow the user to = browse the=20 inventory of any series as well as read the series descriptions. = Therefore, the=20 user interface will incorporate the traditional methods of performing = archival=20 research that maintain the context in which documents were created.=20

Much has already been learned from initial focus groups about what=20 modifications are needed to make the end-user interface more user = friendly.=20 Users commented that some terminology was not easily understandable. For = instance, they were not sure that "score" correctly expressed the = meaning of the=20 rank order or even needed to be shown. Terms like "augment" and "parse," = common=20 though they are in linguistics, are not clear to end-users. With respect = to the=20 structure of the interface, users want to save queries and reuse them.=20

Users also commented on what should be displayed in the list of = retrieved=20 titles. The initial focus groups discovered that users are confounded by = a=20 system that can retrieve a variable number of documents based on = relevance=20 scores. They tend to want "all" of the retrieved documents; however, = this is=20 probably an artifact of past experience in keyword and Boolean retrieval = systems. Similarly, they expect the use of "not" to be Boolean in its = effects,=20 though this effect is not achieved when retrieval is only on nouns and = noun=20 phrases.=20

Upon completion of the initial version of the end-user interface, CMU = will=20 begin conducting formal user protocol testing to provide concrete data = about how=20 researchers actually approach the user interface, and to make changes as = needed.=20 In focus groups, users will often describe what they think they need, = but in=20 protocol testing they actually want something else. Additional tools and = features will be built and introduced as users become more experienced = with the=20 system.=20

5.0 Potential HELIOS Benefits

The HELIOS project team anticipates that the system will have a = number of=20 potential benefits:=20

6.0 Conclusion

By effectively utilizing imaging, OCR, and natural language = processing=20 technologies, the HELIOS project promises to dramatically transform the = Heinz=20 Archives' services by providing researchers with state-of-the-art = electronic=20 access to archival source materials. The HELIOS project is building a = prototype=20 of the digital archive of the future. Hopefully, it will be one of many = similar=20 projects that will make archival information instantly available to = users across=20 the globe, offering them advanced information retrieval capabilities = that=20 significantly enhance their research activities.=20

Notes

1. Donna Harmon, ed., The Second Text REtrieval = Conference=20 (TREC-2) (Washington, DC: Government Printing Office, 1994).=20

2. D. C. Blair and M. E. Maron, "An Evaluation of = Retrieval=20 Effectiveness for a Full-Text Document-Retrieval System," = Communications of=20 the ACM 28 (March 1985): 289-299.=20

About the Authors

Edward A. Galloway, Heinz Archivist, Carnegie Mellon University, 5000 = Forbes=20 Avenue, Hamburg Hall, Room 1506, Pittsburgh, PA 15213-3890. Internet: eg2d@andrew.cmu.edu.=20

Gabrielle V. Michalek, University Archivist, Carnegie Mellon = University, 4825=20 Frew Street, Pittsburgh, PA 15213-3890. Internet: gm1l@andrew.cmu.edu.=20

About the Journal

The World-Wide Web home page for The Public-Access Computer = Systems=20 Review provides detailed information about the journal and access to = all=20 article files: http://info.lib.uh.edu/pacsr= ev.html.=20

Copyright

This article is Copyright (C) 1995 by Edward A. Galloway and = Gabrielle V.=20 Michalek. All Rights Reserved.=20

The Public-Access Computer Systems Review is Copyright (C) = 1995 by the=20 University Libraries, University of Houston. All Rights Reserved.=20

Copying is permitted for noncommercial, educational use by academic = computer=20 centers, individual scholars, and libraries. This message must appear on = all=20 copied material. All commercial use requires permission. =

------=_NextPart_000_000B_01CA0C5E.F4E57CE0 Content-Type: image/gif Content-Transfer-Encoding: base64 Content-Location: http://epress.lib.uh.edu/pr/v6/n4/pacsrev.gif R0lGODdh7wGkALMAAAAAAAAAqwCrAACrq6sAAKsAq6tXAKurq1dXVwAA/wD/AAD///8AAP8A//// AP///ywAAAAA7wGkAEAE/zAFSaet+OrMt+/gJ4bkaJboqabs6rbwK8f0bNf4reckv/++IHAoLBKP xiRyqWwyZ87oU0qdWqvYqzbL3Y66YG94LC6Tz+b0uZlou9/w+OU9ocML7fp8gqdU+ngBDXUFFhuD ggF8hgmFeIWJkIOQjo2KlAGPi46Zl56cjqChnQWgmaMEpwGpBIWppY6vsqqlq7GZs7ClrQWzvL26 rL22vMWwrK3GBMm9yczMwM/RyQ3L09XA0QzN1a0M3s3b4svbBNXlDNjf3wTj6+3t6O3n6+4M9fHz 9+n78fj7AAH6CwgwCRs4deLksdNAYSOGcyQ03NOm0EIJfCQyGmSpQqKOFv9DYkxE6mEhRJxIdiJZ y5RLWqJwkXIls1arU6NaDisAYCesVape6dpVE9qwX8JkZfu1q2k0Yk2tPXPWrKo0qcDKdaMK7Vsz c+16lQsbTmwBeWLDzlsG7929tvTyvb13Np3cfvfirqNH8F/ffQaZOLyohzDCi4gnNPSTMcTEBBwD 5VkkQXKgQIssaJpUMuXMyiVX1vqUMiYlTq+A2uyUujVMobZuDVUNC0BQV6V6qr5ZLPbOpLiXHovm tPdV4sW4Tq2qdVk3q2CdS41eVu27s2u91sPO7hz3tnTFia9bF29ev3P/mu8b2KDDhO8P+zFMf06g xX003FcUeSF/QxxlApL/IiQhgtKAlnTSCCYvkeKgaaI96BJql4Symk6n8GLLbbpsKMxvHgb3oXEi LgWVcciciFxUwi3XHHVgTVMddr1gYx03YcljV3fSrYMdPdyxk9ddbqWnl0Dpqadke0gMxhhibnCk UB8PQfbAA/Nd+QAgCVyJkZYPZWSRHpgZmBlogmgUgJaKWKIllpkgQmCEnbBJi4NstnQnTT/p6ads q7y5JZ9A+aITUT8pFQwuRx3j24irGGOiNcIV8CaLvWgpzYrSjAWONdtkZVYr6pBjalvcfQePpQ+c E09dcIH3ql/l8YVkQLYSNBBBTBbhJByLFWZHlG08lsdEE4FJiBsTgBkm/5sSKFuZRiYBOKcgdpop YCF5XmInt1fieaUj3bL6QJ3hYvKKoOOqxu67QvWUZ3DgPpALK5omGpygsaUi6DP9ZgPMmyOyyIym 0miJjZZNMQyPll4RALFZCEt8pTfmKByWpj5C7FZcHN/lqpK63kpyQU0KNhh8zEIpWcvzCbtHgJBR IJliGYkZ5iY2UyZgJ4gkiAmBldBZmoIY5uQgnbABxZoisOXEZyi/hEhbhxryZhOiSD1KjNUkNkrc 2MuouNzZXFn13HQ0omPuqOOApZWP43iXI6o6BhkkkfWgt+t5J//VKxFOCiuf4THPx2WxheXXX5th SmnBJGGW2SbRaspJ2f/QKx0YUoQvhS76TKo9vXXpwviW4eo77VZUUyEuCqLWVUu6KTJEYTXNibfL yMyLnz4z1qhntY3x8XajU3dba8n6zzlHgidePA2Y97fJ6g0ehBy/JsTysAm9nGbOGGTA5eT/hWnS Hx39x8dmPn9OiZsMS/jJg/inlFq5tAVFOmw3uRBNtOa6RhWKUZlKlzAuxZMPFcdRWnvgilo0Qbad bSvXmIrbQIWx4qkFHq6KDl3kkirtkHAu5EmSP/iSK/X443oB0R4QfvWGxRTLDVSqEuJkdhEpYYBm llETRh6nvs8NSE6W+JwhPJGJdpmiXAxzomjMda564WZNF6vJTOpXinn/WayKOOkiwyyEmwg6TIxs Skq+JjWMKNqrGGvMF3AOdjFBseorxHNGN7Yij26wwzoc/BHeMOad8aTnLrCq3gv5oQ/qnQyGS0qZ exSSEB9Ssnw8fNKTCrMs+0yLfEHUjCW0BRkmagJNpuzctjzBGUykK3+hc1BqknZFAPoPGP1r3VB8 UkvZAcdERwkYU2w3HGf0q2BXOU7agpfHUG1weTe6m6tGRo7v2IN5c7mLXLSZJIKUZx/fDJwMeXDJ OBirfHfgHg7rQwEp3cxyOiTi5YJ2Sk1cxn31JNonhFahVR4tlqRIxdLUVRNA5WJrUZPJQX0Tldgh c1GtAEDXGgWNFGFF/5nKhI5GpfKcFyEHRhuM5ljEcRYgoVBIIPwHNnNkD/SEJ3AwRZmvVEbD+awT SoZTjCbzU5GF3KyU6rNZBkCJICMWTU6PuOfQlqrP/MHyaU41XdRiMkunzW41WdMXvRQVO4jmLne2 +1rZKmqwj7ZomWmLWB65MhY/fkpIX3kHS0v4HUWqdC/bKVl4zOHCvvaNV5JUg2DRQNjBGrawiD2s YqmQ2MYu9rGOjSxkJytZLFSWspi9rGYzy9nNVtaznQ0taEcr2tKSlpymTe1pVcva1bq2s6+NbWtn K9va0lYwt7WtbnPL2936liK9De5vhyvc4mr2IDSsgw3lo8nG9VS5ev+gWWGC1qb8bE4/HqGnISyz XYxogiWkiVNoIGQKmAw0lhe6UwBxwl4CEpCiruNNwHzSVRL1S5gCS06llKHRpZTjKzEKaVaS0Va2 PLOad2vON+JSSLnNapuHzOZcjvTSHbn0ZOPcQXxoaCzDSW6dztUpzp47GQwYcVr1/ORlFlTd6yaI lUC7H4MAWppR4K9/rjlUGAGFta2qLqv+40VPBFbf+RoDKvuNFNk4VUFi+q7AIhyVg40nN7PQLUfk YZ5drbmj6NWDLwP55kAgWT2SZRgHNYVZDs1pU6DeZ6fqmydG4tzOy33kI/n0WREzoxJIrAR0Mh6o oCd0Y6H8aWqoYI3/ha7GCpx8qFBfjdR92Siw2hy5okquIAWV47upYOW/bItYlU1lPA+GMITTa/CE seyXfNi1b7NaD0BeDUn2BNZXK7thmhHXXP+07AKP8PUGjOriVWpLicd2cSWO5k+j3ZhC7DWFLdNr 6GjfElFTYyikFUqvqyaZdphKBgIvmtFOb9QqUja1gfEID5FG2YQgXCn0sKlqWhmpm/6oS5n3se+/ 9BswtyZcrsvJgV67rA3S4sC7GFY5aEFLJNvNTP2ayCZE1I9+W0LXwjnRrTdZtX7lIlSfsFiLSxUQ ja0Yo7/C9cVfpHGMYkygvXaRRYeJtUUEYzJZWy5zOi4cXmBhVwMI/9aNiT1sXOKAGDY5xrGh/1w8 /OamrAN3vTPfYNflhM+Ib+g9PyS8MVg8l83shHAwWWRcNstngh6umU2YPOxgDPtKuBh2jjuRiuSy Fyx5wkULrbxdvDA5q4DCQIud4lI3adcYAeCwM8qRgrnLuTLG5XOqcKzn3nBYNTRV9IuVQ+nfyDkD LrUOzueI9O24UtIvpshXowdW94ZhrQE+00kSXJ3w+WlPq8SYx3DyWNQK4pzJR885f3dAKtEuP0UD Ghrb7zSLVhpNcFzQk1P1NdzmyS6v2Euxia1gTyER+PfLX53LCMAadZsH1XIWdkkFrg6+ct7ilp7i kfSke5H6Ios0Zv/Azf4eVkcDaRYHPLV7OjQshgFc1II+5PMzeXAzLPFJJJEg1SIJ+DQgdCd4TsUg /2NeUlN9AtUahBJAOiZf+uJoBWUucJRzlJY6lAIwFXVGyWRWF4VugQRqbtNgpUIjgDRqL2JC9gdh 3DQyEaZCSSJmAbEr/xeAUEBwvKZ17ORrVSIl/fFmnSRdJtY+IFFUDPgzmJFKTGQm75J3TpQu5oIu TjMvaChQb1MvOGFzKhgo4wIuuZN9/fJzYZVyW7KCMyeDF0NzlKeHz4AwKohyCQQqNrhH0rGIHXUj aLFg5EBNkIh/gyRhzxNmU/dXgBNTtEc4NDWAt4eAwsJTNOMHmBH/Z8oXRGPyCMrnPgdyCdWSRKuE RXA3ca70SiXHLk9kcuvCL7noRh6ni1/kce41bhSniw3zL4BYeHf4Rsn4eMNIeTJyMclAjeoWHXYj UtXRQQeGHTdiVwkGSK3Hf29Ba0ZIjlD3f4ITcNuzYbgHPjXjB+60EERkiru3OBBIYm0Xi6tofCqG YgUyRc4GaKMDClOFOre0J7XkP7fBOojmYwiFgl/1fTdHHPhFTBZlfhh1bggGN8y0QXzkkXKFUiMj SPhQifNWhPkXdd1EMv9ma7XXJN0Dj0DlQ8JyTjzEEfO4LO+0Z6RkT8bniks0PxYYY0aJEk/VIOg1 gqrQlDl2fQJE/21hdG0HRF8uqH3Z0BOKYhS/5GRToWQZNYPvh37OgI3TkY3OVB1u4yOr8lLhWJJF eEJ4YVf7xjd5dWEw6YnEtZfG1Zd8+ZdXAJiC6ZeEOZiG+QKHmZiFuZiKaZiNyZiQ+ZiS2VuTWZmR eZmWeVyYuZmZyZme6VmdGZqfOZqiGQWlSZqoeZqqeXWp2Zqr+ZquuQKwOZuxSZuVaZu1mZu4eZu6 2Zu76ZukiVyhCCUOESCL8WES0XtCdTkXcYroEzkecIqlIDQUuGIcyBlI6Wz/ZBrM5icK2ZB6Il9Z Y21CVpEoyBSxMWlb430UmYeaVm5iiVbEIWrAAzfhMGVAGDcjI/+JJRRriGSJ/mk9Aqok/beOMYlr A8dc9LGcNrl7H6BDp0hdQDVsCzJEPrM5lZMZJ7GBo5RKHIhe6kU60pdtOsYnAKRVU9MbteFtuFQ7 xSFMRQGW7blfUcFpZyMjW4FBbGEWarNu2giJH5SS/alIzANhdyk9f0MkUudXgHWgAjeTmnQ4LdOg dIaF4XOPqOg+7ANx95QgnnNdnhES3Rlo5SV9imZeuWQ6tsET8VV9F+JAwZGe3sdQ9/JVc/RLZ1VW NRiWy8SRwoMjFpQ8HtmD7mAqOxIr/OAWWnak5pErL4krnMgATAgDoMh1C3pONzViyElUFKF7kcA+ 1TUtL9Y+K/b/Z/Mziwwiix9KkB/onYt2kNPHYyd6NRuyE7ZBgi5ojAY0TNx2IvclbmSlkXlKFUCI iJ9iFd3obtHUbt0RMdoBe+RgD0HSahGmF0dijgQaQ+w4Q6DYdTDTdQVocBHBU3fWe8xZJWOibBUK Y6gahmDYGeRFpkvDhiI6IVLDlOLZq+PJS7sEaZUmHJOWFF4Dg3ZaVpvGkU3RHMxRllYmZcsaUjyi n3lVidNzUtmhF5iIPTCljpOKmAoKLLfnrZbEQx/BATsZIHiWGcVHICv7XQEJi/JDCxU6YwT5bAkp ciNqP9W2s1CDoo7ypin4s13jlWbzojtHNr7Kp53GsIHEiKTG/yPa+G5lMULzR0jhiI5KaGEqWZdJ CFMdywID+D0+xT0vU4AKSGeQ81PnMzOg9LLHBwn8uAn/kRLZqTQbyJ1QRW3ZJoJVOWQdiGjpyatE Zl8RxEZhAynHYX6bZqPpVyNsQWDIKmXPJBb0h2BuYWUpxax1UySwIjfpyG9dO6DqKKnbSk6VumYG lzgKUwfO0ixeQkVWmmI+iaFzEhkd53GW812nAXd74i0ZiIs6y14I2SdoZDqvyiEmyijxlbyFS1HD gSnIkWkXRTDVKIMIa5YfBLlsYSPHw4MQmxaJBEg7oirukELYRK2wNiScWKBNqpcyWalBxTiJIYXW 5SyY1CVeov8Bxpk+ECdKyzaBb7N29VO37fo2/ZQSauidVUVeyZuLvLuiXiUi7fUU3scUFfyrWTmj S4Z4Gzxz8UmswjqJgMoO3uhWaplg5kI3eeVHV0aXLAV1E0YysRZTXyubNWU44YpTNeOteNC6/REt +SsR0OK67aMs0km7CyeUFScg3SJ3ZJhxZ0iGpPCHcnguC6UTjDcohAdzrcFy9ZJ4e3iMuegUmhJ2 gbeHbWgxVXFWkcLBwmoxMShHhPhFxfM2bIEwn6d67Yd0efwAQhIy56B6bcExo6d6dJm+1Tq66htJ TtqO7/grkmFJ9kgfDyeFypIIFzfEF5qKc4u/GcdEIME/A1z/d3YHRmN0jIc3h2lKvKpQxheCu72o xYcCh0+RLwHjG/JSP8bAgn06MNCoc3LkDA5TyK1yx3qceoHodHvYx6tnL95xeXT8DqDnPMSMvue4 pJpIw6WrYR9LLFGyQz00yYPwc4/DLp4sLdgiKJFxXZSRxCnLGfxidlPsi09cRf/SyrhrXrF8KVpZ xffcRryoC/QsFO6HC2Poy/wiKUeBO/oFL76D0HMozATjfsJIzP8STYJSzWuR0Rv9L4Gc0dvRSIlq HtOqsRi2zWimTgj4yHUGM1E6PiRwM8ZpYsMmSvg0YnCrREVpGvwEot4ZbTnrqhRCS1XZXjAhG4ay ngvJqxb8/1BhSQxamV8NlbhHO4OgpogElnmeJ2V+lBZR1pam4ipctn/nu6RG2pImHU6duD2f6NLp lLqKo2ab9Gu+xqU0naUpy85eCBr8tHz5xIGCtoE1O0t4G0BXjK8tcVAL5V5Gi8HdtqtGa7Qz+J4a aUFToQ7pp72OiDEm9A4mtVL1p8L/KR4puX+iK8OM7L4IutKFo6DeuiwH6ElYitMWOjnDN3yWA7OY c0/+JIuquqpK+XyxRNilw31kFLi00GiHUocUiVUx+m0INdUF67y7w1+6c93mtqwD9tUkfIPgsJbx BtYiHCtCGJcqidamjd4GqtpP6o7iOiy+963KCdPApo901v8+2rLO64o0EegJsjiLqwsK7SLY2wkT qowhhq3PR43gx/2Q2MZLPHcMbpzU0CBRD9SVD72R+nVRNrIVlr1+A3bC+AlIKaQjASoOhXRNpw17 stZCs5bNAIjSVzeAyPkG+OjWPHU+zVmT9V2q6jomHYozXngt4OVs7wJeGni7qPyL5vVyWURFP8eC dEjHTjFkQxHLAoN44zZG+SJ5CeTPavSHjveHwygVx9xB25vm3Hg33B2E5osO1oqoJoVXM5wdp72J jzR1NYwC8DucyXm/9thJjJEzxNeAPROUiF4LKNahNEteC/ImGEeH+JxG86zFfwcTVm7GrpAnf6ch dmSVUxP/1ZoOd+Gmcvg15tEQR2ROg5py0WQen9owIwe2lt5IteL7HSZ5QvqHb12r1iaN1jHeyNzq hMQJhYA+heKMXfa4Hwu4PoRAJv9IIEGpOUfMfKOsCAn8wBonxcULG/msxmKcyoNSvN3uxBDkS6au h2ckc+OuxqkOjZqu7mjs7tVLedQbDR21C9GBYCK05uIdhNSU63dF2kbqaryOyLeS3mYm4wLY520W pVhotplaSoLuEYgeqttS2/ndYhxvSi8W3P9Ei2b6J0lJ1PdClWGzkK2Dnlc5sE7hE3M0KRmJkW/8 xmzFHA6L1V3tbnGDKsxaZR8D2psrFylODy5ul0eY2mxt/3vuXUPKZYCJgxg3HtcAnDkhRrs2HZA6 Pbfsqrv6xIEamJQeiOBHPfZD5rdT5eBRCTbnGRxDhmlw/0BiBZ8H++GWbSOh8qfttu/BQ1LShIgL Fvh/H38VK3UsvkhEj/R/MXt7fgLd6tZb1069lpOcOte0y8k9U+TrXOR5VhoDAtxkire6JPpTpRsH 5Uu7BKfq+WhMHUxSLffWfeE1rwxUDR07Pw6i0pajgkFwVazyAB5vXtYWNtoplc2OBBAlrsiN3wNh O3BUcpMk+2HPTxj6jWfBZ/EttjMoEWzgpTTHxgmtJK+jI5CoYaYhml5R2c8liFXcV6e95EDCENXA JPuzT//Z2G2s69f3IR61nf37+Qn0EEAaI5QV1upmtuquskKvM7vpVFdTNJMAlmN6tuv61vHdTn5g UAhs/GIy4DFRRCKPuSesGA3EConrMduwxrhWWaELo13F4sD3i+0GztX39z2fu+32wjkfILj1eDGC wL+9PUE/xD6+PMG8gr5Bx8ZFysZJyD7LxcdNAE5B0E1QxkdSy9LSUVVQ1kfQi1NLjVfXxwtYComK ApBd3QyOYIoJiREQ4pYOjBbkFZHlERaWlJUenmsda21sJaHuoaCooCsk8nKjn6z0dSz0LbAqsrV4 N7Z2Oy57sa80ODz/fgH91SH4j1BBNIXoIDKUSKGhQX7/Gh6CFCmTI0aiMjYi1ZFjqFOoVC1Slaok gVYoGamapQGDIFi+UM7CUAvWS1e3lqHE9QvmLw87RwgjMbRoNBfRjEqTts0plKdRuUmlOtVqVaxX tWblutVrV7BfxYYlO9ZsWbRn1WpL23btW7dx4c6VW5fuXbt58e7V66QvX8B/BQcmPNhwYcSHFY9d 3DjxY8eRIU+WXJkyXcuZL2vmvNlzZ9Cfp4YmLbr0adOpUa/mwdq1ativZcemrXj27dq5ce/W3Zv3 b9/BgQ8XDrs4ceTHlSdnrnX58+bRoU+XPoP69erYtWdnvd079+/hwR8WX368efTnu6Znr779e/dk 4sOn/z/ffvH6+e/v19+7P38A/xOQvAALHPBAA9dLEEEGF3SQmwYjfHBCCXGokEIMLzxQwww75JC+ wMARkYhvvlFnHSW0MCKfe9pxkZ57vKBhHnXqsdEM+dBoR4wd25CDDYMGYijIheqAJBE7KkrSoEAk coQhTqL8RBI/MImIo4ymRGUjLkXi4yKRwowlFTJTUkklCkrBxZZHaPIlpltQ0qWmN+XMBZg7X4rz AxE4ECrOEJpZqk8TqCHBUKZO2CvEcIYQZwhyyOkGqnyOYDGHSNO5ptJ8eOSxijX2qWIMHdGwQx9/ 2EhoSILeCJJJVptsSCEoD6qiIouc1CNXitwAMw9PLv+iiEpNxvwoFV9F8bXMk8hk1iU5aWFF2lfm lJPOO+vUaZdhuH0JGZ64fUYZcosiSgRECVWKhaQYWDSuEeOFB50S0Uni3tYsJVWKe2q0QdSAaJQi nkICAviOO7jIYlUihSzyD0wA4WNJiGV945KJD4FIok52/bWUSsrcCMyQhPXyTDGZFdmkWliBtmVX flETlFluyqUlnvS0JecN9BSGz1sADfpPPE9IIamkEE1UhXf5AuebextdgkQRJQXiRHPUsfocgu/p tAZPu/jxjoXrWRVIMwAiW2JXEVq1yIk3nlVujGRVUmOMKIGoblI2CRmkjrREuVhTnM0kpZXNNOWC mMv/dClNnnPRZXKc0syT0HB7KsECECxY5vNDyzW30HaXhoapptuKN+pxvLH3dSUqRWeKFlccGEUw wp6hUoJJhQPIUEf9PeyG3y6eSLiPLHj5AgCYuBJACr5b2bgjzvJKYrMnadmSJSm82FEM/97wwVXC 6c1U1pxpfT9l6pbccDkAQeeia4K/c3WBJnrpo/jvIHW8jOgbTHCdjGDHOnGoyEWfkhQPaqSFgMBg YRL0XRba8AYgGWx4/9Cg8QoGK1PBLW4HeZKSaDWsvFmpesuaVcR+9TFMXGR7J1FWSAQ3vvCVLxU0 6YkrcPYT92GLTRuYX5yOkbNteUB+QDMKupbyxEH5/28FAETL6gTYhCDQLgla5JdftHaDNIShdv3w lwUnGMHemU1tClPVqBKiqrYhj20glFWtRjhCOpxBhr06YSRq2KUo2fCFemQh4Eh2MhyKrBUtUx/l bnImWuwJJjkZ4p5q9hPN+USJeMqf6DhQtCgq5U/tKt3/0sIoEaXodVn0SysTeDUf4GAKnbqdjcB2 S1T1jg1y2KAa1IYGhTlMmIRABMXiprdWIamFSSqEJ3Z1RymBLHx1q0iwPEKSwoUpZSVhFjZPgjhI tsxNjbOcSR6pPnTSzydL3OQyerG5b6HLBfNUgQuUJo1SKuqU8JKaEAj4NATy65U+yBQ7dHlL2e1I gv+o0kfwfKdGXuKhjXH8INz+wEHmsc1ie8PDkfjowkhs6XrXROEND4lIwCkSWc0C5/hEojj28TB9 lZvk5NaURD3hwnPlGhegMDe6zvXvBPlcF+r26TQrjohF/zSHDGhnwBbREgw38KUBP9WPfuWSgjki gwWrmjBh4gMgxHwbCeVowjoqj1ZPKqGTKrJHi2AvkIUMHPhmiMMvqfSbOISpmX54pptp6yXe0uTO emY5oATjiKE83SeFhpShCpV/VMSMo/ppWXvVC4sKtNoTnrqEGeFSdzfKke/4cTaA/UhUAGOYB5O5 1ospcw6eOGYh0Eq3uW0sE8pKod1Y6EzC5dAj2RT/rsq4GYtFXotyMYXkD4coU3HlQmd18lkSiSI0 deEJf5Ed2lHuSVR3HVV1VxxRQaOQ0CZ41i/5+KwDY2TQGMGjbGhTFQVPu8E2ipWixSOrRvVmB2fi 8Y64wuNuo5SxjZZUS1RCBXD/tokpNWuln2CpXr83J+cy12XXiknNDJunEvQip/RjbLlG2b8+GSq7 K1Ba6ShLlqQO9F4EVKUYF4gveskHq7EMw7zMZo8byZcMv1vbGxHWWgouT44LUZ7E+JgltYYMmQ45 ppX4Fgi4Tom3JzXwRzTSvQoPDmZiRlxMKBmumX24kuokIppJEN0mouADO+1uMSCbDEFxNxmJejFc /1anWXZEralaZFFBOyvQzz5QHqA6qC31UVoNEim/aUPIMDEKJUw3OaMe1ejc3trWSzTJen3D0pYO GYptunRMigRnmPtaU1g8Dk1A9LBMxHXJd3rgnUTc3Cd7uuc7S5Z/+ewzWDBb3lUqMKC2M+COt4bV dzy6R6WdBw0qSMEgt9aNli4rfzmoCFvtAdN5Y+tEoMwxTZy71CRLVslC5j1Eiu+4EWYZysinpjyY mWaQdORz7TREwtbvTbu+E7iKArp1pXh0pFwaeMN7F1SSt8YFpVpBH3BxjGdc4xevwcVvh6MofKpF jgZmPDau8bQNySDAC8DFXWsHjp8B48+02PPInf/WJMUcrtDjCG+bR02+EU5wlTAZRQy5ki4R4OQF uLiY9dpqN62PTdv6YU88HFg/ZZJz5zIxTzmJZ4Z30hjsYpp4A+i6GEPqgEbguKYw3iKX/wDjYOOa BYHMVWBiOwYzZ0PGXfWPqso85sOcu+AfMO6XE/ghfMi4rnDV1k9IxMogjebQvZzuw9EVpRg3Hyg4 b9yWhXPfNokZYj+s4fkdFol72qlQDL71rgebnntGmhT1CXF+JpV1OXZReuX+AHbMPQxxTwDfsSrk qfrOHkMuQ8sHnwa/uxELlBar8x+w3zzMXEi1/cf05oakjQsYrdADJCHf2m4oD1ekYDZFXpX++VX/ PMLl8T9J1OlUU5nY5N90wlk8b2o/Bmg61tuZEQBAsRsd7NKufHK42yuLiFulc9A9L2CHSwEC45sl K3g7phM+qNA4heqdSsERDczAubO+B/gClIODuUtBw5sDlxsSlKs5JRmhpttAjvO+iUnBYdE4XOkD HWSEjduIjWMpbLLB6yu6wxkzI2y6Vpi5jIOFjEMfJ4zCXXjCCuDBK9w4negAlCMBjAvAJ4wz2kNA fJoss6si3RMRGvsBJqix5Cs+jzsRp8q4dGi8duDABjA+Sls0FWw7IySbwuvDwzODxjNBPNA+MUBE G4S5F2SejYq57Js5hfBBSERESbTB5nm/B6CI/7nTRF9pOsabvw08Fm2iRFE8upSYOVbgPJfIOA3A Qip8RVUEwyOkAM7DhS+siVvUE4ybny/MwouzxQcoQHVpsXoiQzPEPaTaPaj5J3s5tCYwhycQvn/Z uzikgV60xpNrO9JioL57QbByoxJsQT9ARDcIREW0xEZkK7KpuUgEv06UiEu8G+0bFk/YRkmMwpDR RyxJQgqLxPmjvzJRxZbhPJSYRRu8FlYERlrYxV98iV2kRUDBRwG0wSdyolAqItGxPVNSxvGyIqiB mmxwJRnIRneoQ+LTxjt8wSkIrTJ4NAvSw646xLjjkUC0voFAx0a0vj1QxH84ueqBRCfxuybhvP9k +YRLjKbCQ7VGAABBiEFL0DhvGoSSIEoyIxP4qwWDTEiYkMSnnL8LeEhghMhgDJdbBMb3OUuDk8Vh JAZBSTFugaJpmL1SKra1SMNWuiw3BC02VEkbs8mLk523+8brGxhutDviwQIN3CWygcPr6wLtO0dw TMRBlMzKXETLvEznUybaikTfMrxBBMhNJAmXazyNqUFV/MRN3MBN+Lz5C8XrCyRiWUWvHEjU/DxU qMGvXM3d1MpgdIVezJmAzEMmvIVfDMuyjJMvHAYw3IDlJJSaORoVSJfpZMCy88izA6hjmxpYapS9 hAGN+6ySjMPxBD5R2cZTuargUczJ3L4qoMj/2DREzDQ8mozPzCy8jWLEGowVFqRMHhyEpbuIbfS8 KRxNI0xIelu1UvA7lSBI2py/rNRE39zEXDBI5ITQ3+RFipyFX0zOAQ2B5xQqumw4pgAvu+QKkITA 3eu9gTqvV4KQ3bmG4+sxIOOl0sqlgHGY/JKDOri0PBo3dAPSErIVE4ImUPMD2vKyBku/7JHNVFOp Jw2zRCqzldiwRqqcXQyiDdhSIMoWxeK6EaWz7FKX00FGYXOxM6wsvFzRF2kdvrwxccAadhA5u7OR WkqbHoGHgWCoPEAbNfLTttG2l4MYf1CrZxo18mNHA3MEB9MSCHMwBdstoTMWJjUZw/kS97Mw/x1a hZ/QvyH61IMMzMjZmXiKrup6pz9ZrMQ6mhRwPYw0AT15S2dYmhM9izRUtjfVzgiERnKQqn+xDsaU kR9jNHr4mm4EPIBwqDmYtIaZoyILN3ck1FmBMOvJMnTzLXfrR2rF1LkCpGMBifgbs70SvTOBrsaJ FkfaTVijqTbbKf2BJ2X4GToLGug8iled1WlIRgfMPWRLUTb0vWh8RoFq01aaoBbRkX3RHYVtg0gL vH8AVDjy0WKSnu67FSH1L0+DEu/D1pBCBQg7N5QiqSkVn1S7ygrbIfYxiZo6n8D6HPZRn/vTHPrB GXv1ujvDSGiwpxLlyIfb12Vc03/SIlVqQ/9B+0sbU6+uQT5g1VM12CpTcdgcfZVCmL7jea3+Wisj ZZIAg7zcGlKMUKGPQj9qFRPuyaE/6p5/RC5WU0LFiQk0WTMtlRbGcdedQJ9byxygkZkoKh17WjEU UAqiapdaxYpbhcDY0Rc3pYIIHA2R25eXjFrGXKhR6ZE3CpuUM55LWzKJxUFotbk6UrInO0qM4RuH WD+NcNKiQ78EVTUxmTdyrbdzRVlfaBMNixOayqTCIpTq+rru6rUMMIbZWwqlKVP/GVzGkBc3JYLN Cgct2hoXoaWRa6/b6TGXlCqvesmUs5GIKrJ+wMeqZRVvawgBe4hprRu/CbWMkbx2KyT4PEL/1UVb +LXN0ZQwSxAu0WMTlYGcmEkJI4ocTJo6dvKpOrufnvK/jUQ4KLJOo8JONES77vyzZZMd9lLcCtQx voSHpk1aGr0qPDUygbBcf4jMN+C4iLLaif02i62oiuGYaQ3bksqyauWEeJS/1CzFIgxZZMlKTTUF loIpaJkcaoEc00On6KoT3UWsIvK1Q1G4XzuB0yFePdPXuXjANe1OafzXw42aBJolrjlYHVO0q+JT 5tvRxmybcUwyOKg0i4otoQw3ilnhanUSBMMIZ3Ie04WE1DwE04Q3psyrHS4uKTXZ0GOJlwW41MMp 9BHi3N1C+/mk3XWsX7Oz4AXcyKKGe5qi/zSFsVTaPQI6kUcB2EeZwAjyCx87yXnxFIZCGCAzG9Tq AsBsvBAywReUSv/USVrOOaIMhBWURBN8svDDmyZdBKPMzdpkhHvUxyGkYVE8xUxAud0sTWeOxbcF LNql5mnuCXERrHbKOqMQCqE4YjJMipzyW9szXqqoYge2LAzU4mx4AmsbOa8pZTqlB4FQLSyoKl9K Gx5duhEePEzkScnUA17W5UbURDzuxBSE5X9+Pzu+EgQTZp3r5RrqRIBUZlHQTYJE0N6UUDlRxa9U JH3jsE+FHLu9Kf+tNZmVVwHerqU4naC6SDNlAXP+ijXtBnMotMtqUxmT05Dj0+o1rTTiw/9t6+CH 2jZZDuHBA0eFBgSMa2hZDkLYFGihtOXDw5uPhTeCPkLTFN0lDD9lXtAjrGhl3kYgxOhlPkFrNjPT 65nMSTP/bTPV2zWWhoZYhb2gEVM5K6pJ5siZjgoHbtFkW7sJnlMUEaMw4qylVah59mmtOgPmCzyW izSA/kN/LkeitExA9OfYWsSeBEWAhDBZjjcC201KyOUowU0um8Ip8WgHVaQCheZ9W9crZRxb+6Ek tq5cq66f+oXcbmlgWzEm4ltha8Ap5tfCzek1nLgcQ2wYoTvnngHL9el6Tk+DKYSqej6pPjz93MkN TAh1fMzs/oMbxEk99mzra4TxllCnfGj/dvs8iM7jiram19TEGkwFB4W/r4QkKozCag69kIYWt7Xt 9UGsLTVpXVM9nxoXMkXA2jtAQ7lkmc5kP8PLSflXFu1L3lPcN31DeD7lMUAjoCYja7vcxsxsf46+ +gTvzORszGbNWo6+3szqZElNv1tKXflMq3zv1xxr/o5KiibQsOZxjwbrsMbvc/WrURDifOvUHmry IvYpjdRt2GNpJoKGBy8qshvuvt6KvxZYCnwl511DnTbs7pwXnD5lYF3YLw7W6ruv0wJUZ91JEOK+ 8j2IziSrSNjaFb6yYfnYdEMW1DW1dwPktE3QqcyhcQ29ektyRqIkAiditwwGm+K1TlJw/wY/RhQj Ub6WcGOjcORe0aG1qsL2YlT+wGkjrXemh8p15X5wWOobHoYJpu6Wo1iBpufhtIxi1DpfIWI5ygWL 4WgCKXcbmdOdN3kLZEVn9A1TBbeFE2aH2/9rV1R1V26J9D1D4JolAWwHJU03gS13Ck8H2Khhgl4V Zd5ThzXoYipAoxDkQxw17LKxgjMinopCalhuVkKlc2n13Kwlt/PbiPQFJGdaMAbz1qTzY0o1KdBz 3bVVLrj1IWoeLCBupBIYVXfV9q8DHQOWJ9GhBnIW7igm7rB4QJFEAmdcXpLkMSdId6hwoAiKNmhT PnofmFSJ5Yd17KudcxG6KCWTVkQlpP+vfbcVErpiFzUuybI7PnpCL9k/1lT8AziRKCfqoniZMeIn J8ZAKUAEfGJMX4ETw/IF9tmPJC9d7acKx/Auhud5n15vrIcbXXsji3ujJuqVU7LjkdrwzVqrPiFf V6bRTlQYPl2sXu+UWjVhwVSUOPSGX/QxG6fGV65LUjOTxpbFErFsMSL4eTOtt+tQGsOdpVVOv8tN tiKiBWVy93JPRv2BfdwYnbYwtiAf6RrjUa2BwL45MuFu+/mKpbnPZaajDNLRfTx1u1ZB+jIti99S 5KuSpT+B3FKW1T+fEKKKVz1sAQp1agaEC15xhixy/nxMZmA1JfsH/mRloxQXpZcv8iL/HXtcolZW yR2yVs6lZV2b78X9rq11YCbSi/Gj4we1iulWCCiFhBmItFjunjEBctylTecpqukYTiAMNyFT1PR0 6+9M7H4Ow8ANi7UiEMmYMY5NX5G5hNqaUSj1qd1au9ZAAiwOk8fmcvmsRq/PiTf8DY7T5XVyAz5+ F/h6MlnBXF/fWB9aISIYYRjholhAQ0BFguBj5STmpGXFZuenhKfgaAUpKUXoKGknRkkoqiAqbAFA Rmypq4XuxgiKr4evCUpHyYWKcYkIyzCLCkzKs8rMyU3zjxB2klCOTXdTFQ32EEGDEdaUE3iXuhf7 udeSFxQ8QxvbvZq9Pj5g3Fzdvzt2/wbOGWRHkp89hgAlEtOQTSGHmDRJ2lQpTEVLq0Zp0igKlKdO qnClAmkq1shUo2ThYrlLFcpkxVrxkkWiFzFmwpQBA4EshEwXzYau8LBinA4XQJZe+4E0iA8i2ZhE LWKjqrZyWapu4QpvHr2w8PaRTcOv7BqBBd/k6feH7Vo/Bw3icVMoj5mKFcNEukTJ0kRBeic5zOQR F8W/HyWZ+nhSpMjHsExeqDB51odSJW+hXMnqVWYTxEaTALZzWVGiL5o9Y01tNdMXsWcDubo0CThy OHArwYplq7mu34xYeVcuOHGxX84yz9f8uRuB0gPCdYsQDUM7iS4uzGRoEUbwlPpK7P/E16L5ThUv CQ45KmPj+JA9jrycsqXnV5or66rfP0NlpM3U2U0U+ETaL67whFMLzBzIwQrQSKgUbDxg00MPTXGT VA3eHCGVVR8OZ4MUVDnB1HFgnSNFF8cpl1w90EGHlozU0RHXPzb2YyMYbeGYEHY92vNQH3sZJt4k gxVGmCOfZBQZZB9R9th9oLFiGUn8pQSTlsXo919onQ04E5amkQaAURoYuOCDba7mWjSwUehCNdmM c9UQeOaQoYUcJtFbFUhoJeJwTrgIhRRgEadocS8uR6M+kNY4qaSVUnqppZliuqmmnXL6qaehgjqq qKWSyqmpqZ6qKqurutoqrK/KGiubrbPaCmutud66q6698vqrr8ECO6ywkBB7bLHIKpsss8s62yy0 Mz47bbTVUnuttdliu60Z2nrLLbjfihsuueNGWi665qa7rrrtsvuqu/G+Ky+989pLb7353ruvvv3y C6y/Af8rMMEDG0xWwQkfvLDCDbPrMMQMSxwxxcRWfPHEGWO8MaYce6wxyB+LLHLIJZN88sQoq2wy yyvHGwEAOw== ------=_NextPart_000_000B_01CA0C5E.F4E57CE0--