<?xml version="1.0" standalone="yes"?>
<Paper uid="P06-4003">
  <Title>LeXFlow: a System for Cross-fertilization of Computational Lexicons</Title>
  <Section position="4" start_page="9" end_page="9" type="metho">
    <SectionTitle>
2 LeXFlow Design and Application
</SectionTitle>
    <Paragraph position="0"> LeXFlow is conceived as a metaphoric extension and adaptation to computational lexicons of XFlow, a framework for the management of document workflows (DW, Marchetti et al., 2005).</Paragraph>
    <Paragraph position="1"> A DW can be seen as a process of cooperative authoring where the document can be the goal of the process or just a side effect of the cooperation. Through a DW, a document life-cycle is tracked and supervised, continually providing control over the actions leading to document compilation In this environment a document travels among agents who essentially carry out the pipeline receive-process-send activity.</Paragraph>
    <Paragraph position="2"> Each lexical entry can be modelled as a document instance (formally represented as an XML representation of the MILE lexical entry), whose behaviour can be formally specified by means of a document workflow type (DWT) where different agents, with clear-cut roles and responsibilities, act over different portions of the same entry by performing different tasks.</Paragraph>
    <Paragraph position="3"> Two types of agents are envisaged: external agents are human or software actors which perform activities dependent from the particular DWT, and internal agents are software actors providing general-purpose activities useful for any DWT and, for this reason, implemented directly into the system. Internal agents perform general functionalities such as creating/converting a document belonging to a particular DWT, populating it with some initial data, duplicating a document to be sent to multiple agents, splitting a document and sending portions of information to different agents, merging duplicated documents coming from multiple agents, aggregating fragments, and finally terminating operations over the document. An external agent executes some processing using the document content and possibly other data, e.g. updates the document inserting the results of the preceding processing, signs the updating and finally sends the document to the next agent(s).</Paragraph>
    <Paragraph position="4"> The state diagram in Figure 1 describes the different states of the document instances. At the starting point of the document life cycle there is a creation phase, in which the system raises a new instance of a document with information  The document instance goes into pending state. When an agent gets the document, it goes into processing state in which the agent compiles the parts under his/her responsibility. If the agent, for some reason, doesn't complete the instance elaboration, he can save the work performed until that moment and the document instance goes into freezing state. If the elaboration is completed (submitted), or cancelled, the instance goes back into pending state, waiting for a new elaboration.</Paragraph>
    <Paragraph position="5"> Borrowing from techniques used in DWs, we have modelled the activity of lexicon management as a set of DWT, where lexical entries move across agents and become dynamically updated.</Paragraph>
  </Section>
  <Section position="5" start_page="9" end_page="10" type="metho">
    <SectionTitle>
3 Lexical Workflow General Architec-
</SectionTitle>
    <Paragraph position="0"> ture As already written, LeXFlow is based on XFlow which is composed of three parts: i) the Agent Environment, i.e. the agents participating to all DWs; ii) the Data, i.e. the DW descriptions plus the documents created by the DW and iii) the Engine. Figure 2 illustrates the architecture of the  The DW environment is the set of human and software agents participating to at least one DW.  The description of a DW can be seen as an extension of the XML document class. A class of documents, created in a DW, shares the schema of their structure, as well as the definition of the procedural rules driving the DWT and the list of the agents attending to it. Therefore, in order to describe a DWT, we need four components: * a schema of the documents involved in the DWT; * the agent roles chart, i.e. the set of the external and internal agents, operating on the document flow. Inside the role chart these agents are organized in roles and groups in order to define who has access to the document. This component constitutes the DW environment; * a document interface description used by external agents to access the documents.</Paragraph>
    <Paragraph position="1"> This component also allows checking access permissions to the document; * a document workflow description defining all the paths that a document can follow in its life-cycle, the activities and policies for each role.</Paragraph>
    <Paragraph position="2"> The document workflow engine constitutes the run-time support for the DW, it implements the internal agents, the support for agents' activities, and some system modules that the external agents have to use to interact with the DW system. Also, the engine is responsible for two kinds of documents useful for each document flow: the documents system logs and the documents system metadata.</Paragraph>
  </Section>
  <Section position="6" start_page="10" end_page="10" type="metho">
    <SectionTitle>
4 The lexicon Augmentation Workflow
</SectionTitle>
    <Paragraph position="0"/>
    <Section position="1" start_page="10" end_page="10" type="sub_section">
      <SectionTitle>
Type
</SectionTitle>
      <Paragraph position="0"> In this section we present a first DWT, called &amp;quot;lexicon augmentation&amp;quot;, for dynamic augmentation of semantic MILE-compliant lexicons. This DWT corresponds to the scenario where an entry of a lexicon A becomes enriched via basically two steps. First, by virtue of being mapped onto a corresponding entry belonging to a lexicon B, the entry (A) inherits the semantic relations available in the mapped entry (B) . Second, by resorting to an automatic application that acquires information about semantic relations from corpora, the acquired relations are integrated into the entry and proposed to the human encoder.</Paragraph>
      <Paragraph position="1"> In order to test the system we considered the Simple/Clips (Ruimy et al., 2003) and ItalWord-Net (Roventini et al., 2003) lexicons. An overall picture of the flow is shown in Figure 3, illustrating the different agents participating to the flow. Rectangles represent human actors over the entries, while the other figures symbolize software agents: ovals are internal agents and octagons external ones. The functionality offered to human agents are: display of MILE-encoded lexical entries, selection of lexical entries, mapping between lexical entries belonging to different lexicons  , automatic calculations of new semantic relations (either automatically derived from corpora and mutually inferred from the mapping) and manual verification of the newly proposed semantic relations.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>