=Paper=
{{Paper
|id=Vol-101/paper-10
|storemode=property
|title=Semantic Word Processing for Content Authors
|pdfUrl=https://ceur-ws.org/Vol-101/Marcelo_Tallis.pdf
|volume=Vol-101
}}
==Semantic Word Processing for Content Authors==
https://ceur-ws.org/Vol-101/Marcelo_Tallis.pdf
Semantic Word Processing for Content Authors
Marcelo Tallis
Teknowledge Corporation.
4640 Admiralty Way
Marina del Rey, CA, USA
email mtallis@teknowledge.com
gested as an alternative method for generating semantic
ABSTRACT annotations. Unfortunately, this technology is only able to
Document authors cannot routinely afford the overhead extract sufficient information to fill in a flat template and
imposed by current semantic annotation tools. Some char- cannot capture the relationship graph that connects the in-
acteristics of their task can be exploited to provide them stances ([5][7]).
with a tool that will reduce the effort required to create Clearly, current markup authoring tools are inadequate for
both the document content and their accompanying seman- the task of routinely authoring content. Fortunately, some
tic annotations. characteristics of this task, as it applies to some authors,
can be exploited to reduce the cost of producing these an-
SemanticWord is such a semantic annotation tool. Seman-
notations. Some of these characteristics are:
ticWord is an environment based in MS Word that inte-
grates content and markup authoring, providing customiza- • The documents to be authored are primarily con-
ble tools that allow simultaneous generation of content and fined to a few topics. In this case it is worthwhile
semantic annotations, an annotation scheme that allows to spend some effort in setting up an environment
annotations to be reused when content is reused, a custom- tailored to these topics. The savings from produc-
izable library of templates containing partially annotated ing multiple documents will more than recoup the
text, and an automatic information extraction system with tailoring cost.
the tools for refining and augmenting its output. • There is a high degree of content reuse. For exam-
ple, different documents include common actors
Keywords and places, share the same context, or update on
SemanticWord, Semantic Annotations, Semantic Web,
previous accounts. This characteristic can be ex-
Markup Authoring Tools, COTS integration.
ploited to reuse the annotations along with the
content.
INTRODUCTION
The vast amount of information contained in the web is SemanticWord is a semantic annotation tool designed with
beyond any individual’s grasp. Unfortunately, its content is this kind of task in mind. Some of the features included in
primarily tailored to human consumption and not suitable SemanticWord are:
for automatic semantic interpretation. The semantic web • An environment that integrates content and
addresses this problem by allowing content to be annotated markup authoring. This environment is based in
with machine understandable semantic descriptions. MS Word, a product that is already familiar to
Although current annotation tools take care of the annota- many authors.
tions syntax and the proper reference and use of ontology • Customizable tools for simultaneous generation of
terms ([4][5][6]), authoring semantic annotations remains a content and semantic annotations.
tedious and expensive process. • An annotation scheme that allows for annotations
While this cost may be affordable to people who author to be reused when content is reused.
web documents sporadically (e.g., a teacher authoring her
• A customizable library of templates containing
homepage) it would be prohibitive to those who author and
partially annotated text. Authors can include tem-
update documents routinely (e.g., an intelligence analyst
plates in their documents to speed up both content
writing intelligence reports).
and annotation production.
Automatic Information Extraction systems have been sug-
• An automatic information extraction system and
the tools for refining and augmenting its output.
SEMANTIC WORD
SemanticWord offers an environment for authoring anno-
tated text documents based in MS Word. Its aim is to re-
duce the burden involved in authoring semantic annota-
tions. Authors are given a familiar and uniform environ- language developed for the Semantic Web that supports the
ment where the creation of content and semantic descrip- definition of machine-readable ontologies and the linking
tions can be freely interleaved. In many case both of them of terms in documents to ontologies.
can be achieved in a single operation. SemanticWord annotations are attached to regions of text,
not to the document as a whole. There are two types of
Overview annotations: instances references and triple bags. An in-
SemanticWord extends MS Word in several dimensions stances reference associates a text region with a “referen-
(see Figure 1). First, MS Word GUI is augmented with cable” instance of a class. Triple bags describe the content
toolbars that support the creation of semantic descriptions of a text region with a collection of triples that follow
(or annotations) that are attached to text regions. The GUI DAML+OIL’s subject-predicate-object model. The subject
is also extended to show these annotations embedded is an instance, the predicate is a property defined in an on-
within the text and to support their direct manipulation tology, and the object can either be an instance or a value.
through mouse gestures. Second, SemanticWord extends SemanticWord Annotations are retained across text
Word’s reach by opening a channel to the Semantic Web. copy/cut and paste operations.
Content from the Semantic Web (both ontology definitions
Figure 2 illustrates a fragment of an annotated document.
and factual descriptions) is brought into SemanticWord to
An instance reference is rendered by enclosing the anno-
compose annotations that are later dumped back into the
tated text between square brackets and with an icon adja-
Semantic Web. Third, SemanticWord extends Word ser-
cent to the closing bracket. A triple bag is rendered by en-
vices by integrating AeroDAML, an automated information
closing the annotated text between square brackets and
extraction system. AeroDAML analyzes and annotates the
displaying a checkbox and a triples table adjacent to the
text of the document as it is being typed, appearing to the
closing bracket. The checkbox allows the user to display or
author as a service analogous to Word’s spelling and
hide the table. To facilitate the handling of heavily anno-
grammar checking. Finally, SemanticWord supports the
tated documents, the text associated to an individual anno-
rapid composition of annotated text through template in-
tation can be highlighted and all annotation marks can be
stantiation.
made invisible.
The above extensions were implemented using standard
An Instance reference icon can be dragged and dropped
Microsoft extensibility technology. Annotations are ren-
over a cell corresponding to the subject or object of a triple.
dered with ActiveX controls that can be placed in a docu-
Cells filled using this method do not store a direct reference
ment, implement their own behavior, control their GUI,
to the dropped instance but rather establish a link with the
and save their internal state. Automatic text analysis is
dragged instance reference. Updating the linked instance
driven by SmartTags technology that supports background
reference to refer to a different instance will alter the triple
parsing and tagging of the document text as it is being
too. This level of indirection improves maintainability.
typed. The rest is supported by an Office COM Add-in that
responds to MS Office/MS Word built-in events (e.g., Triple cells can also be filled by picking instances and
DocumentOpen) and extend Word’s menus and toolbars. properties from special purpose browsers called choosers
The document content is manipulated through Word’s (See Figure 3). Choosers can use the values already stored
COM API. in a triple to constrain the lists of choices offered to the
user. For example, if the subject and object of a row are
Semantic Annotations already filled in then the corresponding property chooser
SemanticWord annotations are based in the DAML+OIL will only show the properties whose domain and range are
language [3]. DAML+OIL is a knowledge representation consistent with those entries. Because SemanticWord does
not enforce consistency, these constraints can be relaxed.
The choosers also provide other filters for constraining the
choices shown. For example, the instance chooser includes
MS
filters for listing only the instances that have already been
Word Document Word referenced in the document. The instance choosers can se-
(Text + Annotations )
lectively list instances corresponding to preexisting seman-
Semantic tic web markup (provided by the Ontology and KB Server)
Annotated
Word or new instances defined in the current document. They
Templates also allow users to create new locally defined instances or
provisional instances (described below), a function that a
user would invoke if the listed choices do not include the
Ontology
and KB
AeroDAML desired instance. SemanticWord does not impose any order
(IE)
Server for filling in table cells, and can persist the state of tables
containing rows with one or more empty cells.
Figure 1. SemanticWord Architecture
Locally defined instances are instances that cannot be ref- mantic annotations that is tightly integrated to MS Word.
erenced from outside the document. Provisional instances Word is the most massively adopted product for authoring
are an artifact to postpone the identification of an instance text documents. SemanticWord includes a set of tools that
that is being used to describe relationships. Ultimately, economize the production of semantic descriptions and
provisional instances must be replaced by references to exploit opportunities for the simultaneous generation of
external or locally defined instances. SemanticWord keeps text and annotations. Two examples of these tools are per-
track of the provisional instances and assists users in re- sonal class toolbars and the cascading class menus, both
placing them. illustrated in Figure 4.
One obstacle that we noticed in other systems when com- Personal Class Toolbars: Personal Class Toolbars consti-
posing a triple is that the role of the instances in the triple tutes a convenient tool for generating both content and an-
cannot be established before examining the definition of notations together with just one mouse click. Users can
the predicate property. For example, determining who is create any number of Personal Class Toolbars, each one of
the subject and who is the object in the relationship be- them tied to a single class. Each personalized class toolbar
tween an employee and her employer depends on how the contains an instance selection combo box and buttons to
property that relates both of them is (arbitrarily) defined. create instance references corresponding to the selected
Assigning an instance to the subject or the object of a triple instance or a new one. If at the time the user creates an
prematurely might preclude the possibility of establishing instance reference the document contains a selected region
the relationship. To avoid this problem in SemanticWord, of text, then the instance reference will be attached to that
the property chooser can optionally list reversed properties. region. If no text is currently selected, then both the “label”
Reversed properties are ordinary properties that assume the of the instance reference will be inserted in the document at
subject and the object of a triple are switched. Reversed the current text insertion point, and the new reference will
properties is only an artifact to add another degree of lib- be associated with the inserted text.
erty in the order in which the triple arguments are filled -- Personal class toolbars save effort when a small percentage
the generated DAML markup switches the subject and ob- of classes or instances account for a substantially larger
ject of a triple when a reversed property was selected. percentage of the instance references that an author will
need.
Taming Annotation Authoring
SemanticWord was conceived with the goal of minimizing Classes Cascading Menu: A cascading class menu in-
the burden involved in authoring semantic annotations. cludes an entry for every named class in the ontology at-
This burden is reduced through several techniques. tached to the document. This menu gives users access to
most of the operations related to ontology classes, includ-
Non Intrusive Annotation Environment ing defining new instances, creating personal class tool-
SemanticWord provides an environment for authoring se- bars, and opening instance choosers. When a user executes
Figure 2. Fragment of an annotated document.
The circular icon containing an I Bar (like that adjacent to “BAGRAM”) references an external instance
from the semantic web. A smiley face icon (like that adjacent to “weapons cache”) references a locally de-
fined instance. The boxed legend below the “weapons cache” instance reference is its tool tip. If the instance
icon in the subject or object column of a table is overlapped by a small arrow in its lower left corner (like the
one in the object column of the first row) then the cell is linked to an instance reference annotation. Modify-
ing or deleting the linked instance reference will affect the triple too. If the instance icon is not overlapped
by a small arrow (like the one the subject column of the first row) then the cell contains a direct reference to
an instance.
any of these functions from this menu, the menu entry cor- Automatic Information Extraction
responding to the selected class is duplicated and placed at SemanticWord integrates an information extraction system
the top of the menu so the user can access it easily the next (IES). Automatic information extraction technology prom-
time that she needs it. The cascading hierarchy is deter- ises to significantly reduce the human overhead involved in
mined by the subclass hierarchy of the ontology. Classes the semantic annotation task. Although this technology has
with multiple superclasses appear in the cascade under each not reached a level of sophistication required to capture
superclass. deep relationships in text ([5][7]), it can provide useful
Direct Manipulation of Annotations: Direct manipulation annotation fragments. The approach taken in Semantic-
of annotations is another method of simplifying the produc- Word is to supply the tools that would allow users to aug-
tion of semantic annotations. In SemanticWord users can ment the annotation provided by an IES.
compose semantic annotations by manipulating other anno- SemanticWord uses AeroDAML, an IES developed at
tations that are placed within the document. For example, Lockheed Martin [7]. AeroDAML processes text and pro-
the subject and object of a triple can be filled by dragging duces DAML markup that relates instances and values to
instance references annotations over the triple. For some Ontology classes and types. AeroDAML relies on a high
users this method is faster and more natural than searching performance commercial information extraction system
for those same instances in instance browsers. called AeroText. The default AeroDAML is based in the
default AeroText which includes “domain independent”
Flexible commitment order extraction rules capable of extracting many proper nouns
Authors should not be forced to follow a strict order in and frequently occurring relations. AeroText and conse-
carrying out the many steps involved in authoring semantic quently AeroDAML can be tailored to particular domains
descriptions. Many of the features that support this princi- through training sessions with annotated corpuses.
ple have been introduced before. These features are sum-
SemanticWord provides an environment for refining and
marized in this section.
augment the result of IESs. We observed that the default
• Elements of a triple can be entered in any order. AeroDAML does a good job at recognizing and categoriz-
Even the determination of which instance is the
subject and which is the object can be postponed
by means of the reversed properties. New in-
stances can be created from the instance choosers
avoiding a disruption of the triple’s composition
process. Unlike other annotation tools, triples are
laid out in a tabular structure rather than in a tree
or other structures that impose a topological de-
pendency among its nodes.
• Consistency is not enforced. A user is free to com-
pose a triple that violates ontology constraints. Property Chooser
The user can make the changes that would fix this
conflict at a time convenient to her. Consistency is
taken into account when filtering suggested
choices for composing a triple, but the user can
deactivate these filters with a single button click.
• Instance identification can be postponed but the
instance can still be used to describe relationships.
This is achieved through the use of provisionary
instances, which can be used wherever definitive
instance can but remind the user of the uncon-
cluded task. SemanticWord will assist users in as-
signing identity to these instances.
Instance Chooser (Object)
Annotation Reuse Figure 3. Property and Instance Choosers.
Annotations are attached to text regions and are going to be The choices correspond to the filling of the property and
reused when those regions are reused. In particular, annota- object columns of the second row of the triples table of
tions are carried over along text cut/copy and paste opera- Figure 2. The listed choices are constrained by the content
tions and when fragments of a document are reused else- of the other cells of the selected triple. These filters can be
where in the same document in other documents based on relaxed by toggling the buttons on the top toolbars. The
the same ontology. Instance chooser also supports the definition of new in-
stances.
ing proper nouns but their classification tends to be overly persists as a (typically quite small) word document.
general. It also fails to recognize most of the relations be- A template may be inserted into a document just like any
tween instances. For example, AeroDAML succeeds in other document. Both the text and annotations of the tem-
classifying Kabul as a Place but failed in finding the more plate are copied into the target document. After insertion,
specific class City, perhaps because there was nothing in the copy can still be subjected to further editing and anno-
the text that might clue AeroDAML about this fact. Se- tating.
manticWord let AeroDAML to recognize and classify
Templates are authored in SemanticWord in template de-
proper nouns but expects the user to refine the classifica-
sign mode. All annotations tools described previously are
tion and to specify their relationships.
also available for annotating templates in template design
SemanticWord drives the information extraction process on mode plus an additional toolbar that includes the template
the fly. As the user types the content of the document, a specific authoring tools described below. We expect that
background thread feeds new or modified text to Aero- non-programmers would be able to author templates.
DAML in paragraph units (roughly), obtains the extracted
Instance Placeholder: An instance placeholder annotates a
entities with their position in the text, and underlines those
region of text that needs be replaced by an instance refer-
text regions with a blue wiggly line. This procedure is car-
ence when the template is used in a document. It also
ried out in a way that resembles Word spelling and gram-
serves as the surrogate for an instance reference, and as
mar checking and is implemented in terms of Microsoft
such, it can participate as the subject or the object of one or
SmartTags technology.
more triples in the template’s triple bags.
The user can examine the extracted entities and convert
An instance placeholder is rendered like an instance refer-
them into instance reference annotations. As part of this
ence annotation but with a different icon. In design mode
conversion the user has the option of refining the extracted
this icon can be dragged over triple tables to compose the
type. Once an extracted entity has been transformed into an
semantic annotations that describe the template. It can also
instance reference it behaves just like a natively created
be dragged over another instance placeholder to specify a
instance reference. In particular, it can be dragged and
co-reference requirement. In instantiation mode, this icon is
dropped onto cells of triple bags to describe the relation-
a drop site for the concrete instance that is going to be
ships that AreoDAML missed.
bound to the instance placeholder.
Annotated Templates When an instance placeholder is bound to an instance ref-
Annotated text templates reduce the amount of work in- erence, the label of the instance reference replaces the tem-
volved in authoring both semantic annotations and docu- plate’s text and all co-referential instance placeholders are
ment content. A template consists of a text fragment anno- bound to that instance.
tated with semantic and template related descriptions, and Optional group: An optional group delimits a region of
Figure 4. Toolbars and Menus.
The last two toolbar rows belong to SemanticWord. The first row contains two juxtaposed personal class toolbars. The
first one is tied to the class “Terrorist Organization” and has selected the instance “al Qaeda”. The second one is tied to
“Country” and has selected “Afghanistan”. Clicking in the Check button will generate both the text and the annotation
corresponding to the selected instance. The other buttons are for defining new instances before inserting their text and
annotation. The last toolbar row has its classes cascading menu opened. This menu provides access to several class
related functions. The most recently chosen classes get added to the top of the menu (like Weapon, Terrorist Organiza-
tion, and Country) for easy access.
text in the template that can be optionally included in the • Markup that is tied to text fragments disappears if
instantiation of the template. The text delimited by an op- the text fragment, or a region containing it, is de-
tional group can contain annotations and other groups. In leted. Generally, this is desirable because the
particular, it can contain instance placeholders. Opting to document’s content no longer supports the state-
delete an optional group from an instantiated template will ment formalized by the deleted annotation.
automatically remove any triples having a cell linked to an Among the difficulties of this approach we found:
instance placeholders within the deleted group.
• If an entity (e.g., a person or place) is mentioned
Repeated group: Like an optional group, repeated group several times within the text, it might be necessary
annotation delimits a region of text and can also contain to duplicate its annotation too.
other groups and annotations. During instantiation the user
can ask that a repeated group be replicated any number of • Some concepts might be implicit or too abstract to
times. Each replication of the group creates its own incar- be located in the text.
nation of the instance placeholders that it contains. When • As changes are made to text within an annotated
the group is replicated, all triples with cells linked to the region – particularly at its boundaries – heuristics
instance placeholders contained in the group are replicated must be used to adjust the boundaries. The use of
as well. paired brackets for rendering these regions keeps
The utility of annotated templates is enhanced by the IES the user informed of the result of these heuristics.
described above. The IES analyses the document and gen- Although SemanticWord is biased toward the attachment
erates instance reference annotations corresponding to the semantic annotations to text, it does not mandate it, open-
concrete entities mentioned in the text. These instance ref- ing a whole spectrum of hybrid compromises. For example,
erences can be dragged over the template instance place authors might choose to attach instance references to text
holders to instantiate the template and generate instantiated but to describe their relationships in a single global triple
triples describing their relationships. bag. Moreover, not even the instance reference annotations
are required because the triples can be filled directly from
ANNOTATING TEXT REGIONS instance choosers. More serious use of SemanticWord will
In SemanticWord, semantic descriptions are distributed be required to weigh the pros and cons of this approach..
throughout the document and attached to text regions that
“support” their content. This is not a requirement for the RELATED WORK
semantic web. Most of the semantic markup authoring Research in semantic annotations is still in its infancy. A
tools reported in the literature do not adopt this practice. number of systems have been developed to date that dem-
The descriptions they produce are associated only with a onstrate different capabilities. However, the approaches
document, not with portions of that document. adopted by these systems do not necessarily compete
We speculate that relating a semantic description to the text against each other but rather address different issues.
that supports it has advantages in terms of annotation au- Ont-O-Mat ([4][5]), one of the first annotation systems to
thoring, reuse, maintenance, and validation. However, we appear, is the concrete implementation of CREAM [4], an
also recognize that this practice might introduce unneces- annotation and content authoring framework conceived for
sary complications. the easy creation of relational metadata (i.e., relations be-
Some of the advantages of attaching semantic descriptions tween instances). Ont-O-Mat includes its own HTML
to text are: document editor for viewing and composing the content of
the document being annotated and an ontology and fact
• Descriptions can be reused if the text is reused. browser for visualizing the markup collected by a crawler
Annotations are carried over along text cut/copy and for authoring the markup that annotates a document.
and paste operations and when document frag- Like SemanticWord, Ont-O-Mat also provides mechanisms
ments are reused in other documents. that simplify the creation of markup, document content, or
• Conformity between the semantic descriptions and both. For example, dragging text from the document editor
the content of the document can more easily be and dropping it on top of a class listed in the ontology and
validated and maintained. fact browser could automatically create an instance of that
• Authors might find it natural to find annotations class with the dragged text filling some property of the
by finding, through familiar text search/scroll created instance (e.g., its name). Similarly, dragging an
mechanisms, the text to which the annotations are instance listed in the ontology and fact browser and drop-
attached. Contrast this with browsing the semantic ping it at some location within the document editor could
markup directly. For example, in SemanticWord insert in that location the text corresponding to the filler of
authors compose triples by dragging around in- some property of the dropped instance and eventually
stance references placed within the text. could attach to that text a hyperlink that describes the in-
stance further. A meta ontology specifies the type of ac-
tions to be carried out through the dragging and dropping CONCLUSIONS
operations. SemanticWord integrates into a widely used COTS product
S-CREAM [5] extends the CREAM framework with an an environment for authoring document content and anno-
information extraction component for the semi-automatic tations. It includes several features intended to minimize
generation of annotations. In S-CREAM manual annotation the cost involved in authoring semantic annotations: cus-
is supported by Ont-O-Mat while automatic information tomizable tools for generating content and annotations si-
extraction is supported by Amilcare [1], an adaptive infor- multaneously, direct manipulation of annotations embed-
mation extraction system (IES). Because the IES is unable ded in the document, reusable annotations, annotated text
to capture relationships in a graph that connects the indi- templates, and an information extraction system including
viduals described in the text, the output of the IES has to be support for refining and augmenting its output.
mapped into a Discourse Representation (dependent on the
domain) before generating a set of markup hypotheses. REFERENCES
This technique is still very rudimentary. [1] Ciravegna, F. Adaptive Information Extraction from
SMORE [6] provides an environment for composing the Text by Rule Induction and Generalisation, Proc. of
content and the inline semantic annotation of web pages, 17th International Joint Conference on Artificial Intel-
email, and other online documents. Like SemanticWord, ligence (IJCAI 2001) , Seattle, August 2001.
SMORE aims to support semantic annotation without dis- [2] Ciravegna, F., Dingli, A., Petrelli, D., and Wilks, Y.,
rupting the document creation process. Toward this end Timely and Non-Intrusive Active Document Annotation
SMORE supports practices like using place holders to de- via Adaptive Information Extraction, in Semantic Au-
fer the final determination of the markup, referencing mul- thoring, Annotation & Knowledge Markup (SAAKM
tiple ontologies that can be brought to bear when the need 2002), ECAI 2002 Workshop, July 22-26, 2002 ,
arises, and extending ontologies if none of the known on- Lyon, France.
tologies fit the user needs. SMORE also integrates several [3] Connolly, D, van Harmelen, F., Horrocks, I, McGuin-
unique capabilities, like the ability to annotate parts of im- ness, D., Patel-Schneider, P., and Stein, L.
ages using SVG, an advance ontology search capability, DAML+OIL (March 2001) ReferenceDescription,
web scraping, and a Semantic Virtual Portal that provides W3C Note 18 December 2001
links to semantically related material http://www.w3.org/TR/daml+oil-reference.
MnM [8] and Melita [2] are environments that streamline [4] Handschuh, S and Staab, A. Authoring and Annotation
the automatic production of semantic annotations using an of Web Pages in CREAM, in Proceedings of the
information extraction system (IES). The process supported WWW2002 - Eleventh International World Wide Web
by these systems comprises several activities, including Conference, Hawaii, USA, May 2002.
manually annotating web pages (for training the IES), [5] Siegfried Handschuh, Stephen Staab, Fabio Ciravegna,
training the IES using the annotated pages, tuning the per- S-CREAM -- Semi-automatic CREAtion of Metadata,
formance of the trained system, and running the IES to in Semantic Authoring, Annotation & Knowledge
automatically annotate a set of pages. MnM implements a Markup (SAAKM 2002), ECAI 2002 Workshop, July
generic process model which is also generic with respect to 22-26, 2002 , Lyon, France.
the specific ontology server and information extraction tool
used. Melita is a demonstration system that seamlessly in- [6] Aditya Kalyanpur, James Hendler, Bijan Parsia, Jenni-
tegrates manual annotation, incremental training, and auto- fer Golbeck, SMORE - Semantic Markup, Ontology,
matic information extraction in a timely and non-intrusive and RDF Editor, available at
way. These systems have demonstrated that is possible to http://www.mindswap.org/papers/SMORE.pdf
highly automate the generation of semantic annotations. [7] Paul Kogut, William Holmes, AeroDAML: Applying
Unfortunately, the scope of these annotations is restricted Information Extraction to Generate DAML Annota-
to only filling in one information template per document. tions from Web Pages , First International Conference
Both systems use Amilcare as their IES. on Knowledge Capture (K-CAP 2001), Workshop on
Among the described annotation tools only SemanticWord Knowledge Markup and Semantic Annotation, Victo-
provides an environment for document authoring and se- ria, B.C. October 21, 2001 AeroDAML.
mantic annotation that extends a COTS product that au- [8] Maria Vargas-Vera, Enrico Motta, John Domingue,
thors have already adopted (MS Word). SemanticWord is Mattia Lanzoni, Arthur Stutt1, and Fabio Ciravegna,
also the only one that associates semantic annotations MnM: Ontology-Driven Tool for Semantic Markup, in
within text regions and consequently facilitates annotation Proceedings of ECAI 2002, July, 2002 , Lyon, France.
reuse and maintainability.