App. C

Appendix C. Explicit Structure of the Hermes Language

Syntax of the Hermes language. This is a complete listing of the options of the Hermes language in BNF format. This is the full advanced version of the language, incorporating all the options for the beginner and intermediate levels as described in Chapter 10. All Capitalized Terms are non-terminals. Underlined terms are literal terminals. (Words in parentheses) are comments. Other terms describe terminals. [Terms in square brackets] are optional. The start symbol is DataList.

--------------language elements----------------------------------------

DataList ::= SimpleDataList | ComputedDataList

Association ::= SimpleAssociation | InputAssociation | ComputedAssociation | Predicate

Predicate ::= Association

InputAssociation ::= LinkType | InputAssociation with their InputAssociation | InputAssociation and InputAssociation | an input association name

CharacterFilter ::= include Character

NumberFilter ::= Counter

BooleanFilter ::= true

ContextFilter ::= view [Counter] DataList | inherit from Context | are inherited by Context

GraphicFilter ::= [immediately] contain Graphic | [immediately] contained in Graphic | Measure [Quantifier] Graphic [in Graphic] | have Attribute is Value | have Attribute is Number

---------------media elements----------------------------------------------

Character ::= SimpleCharacter | ComputedCharacter

SimpleCharacter ::= character string | a Character name

ComputedCharacter ::= substring of Character from Number for Number | Character append Character

Number ::= SimpleNumber | ComputedNumber

SimpleNumber ::= real number | a Number name

Boolean ::= SimpleBoolean | ComputedBoolean

SimpleBoolean ::= true | false | a Boolean name

ComputedBoolean ::= there are Counter DataList | Quantifier DataList Filter | not Boolean | Boolean Connective Boolean | Graphic Measure [Quantifier] Graphic [in Graphic]

Graphic ::= SimpleGraphic | ComputedGraphic

SimpleGraphic ::= polyline | a Graphic name

ComputedGraphic ::= DataList (of type graphic)

Image ::= bitmap image | an Image name

Pen ::= pen sketch | a Pen name

Sound ::= sound segment | a Sound name

Video ::= video segment | a Video name

Animation ::= animation segment | an Animation name

ComputedView ::= DataList arranged in a window | a ComputedView name

-------------------network elements-----------------------------------------

NodeKind ::= a NodeKind name

LinkType ::= a LinkType name

Context ::= a Context name

ResultList ::= name of an evaluated DataList

------------------namable terminology elements-----------------------------

Measure ::= Distance is Counter Units | not Measure | Measure Connective Measure | a Measure name

------------------simple terminology elements----------------------------------------

Connective ::= and (logical) | or (logical)

Combination ::= and (unique sorted union) | and also (intersection) | but not (difference) | or (append) | with (and, indented)

Distance ::= central distance | closest distance | x distance | y distance | z distance

Units ::= inches | feet | cm | meters

Denotational semantics of the Hermes language. The semantics of the Hermes Language was formalized using the notation of Schmidt (1986), based on the denotational semantics of Strachey and Scott. The abstract syntax, semantic algebras, and valuation functions provide a formal specification for the Hermes source code implementation. Each option in the abstract syntax is programmed as an object. The semantic algebras are implemented by these objects, which are given methods corresponding to the operations specified for the algebras. Evaluation methods for the objects correspond very closely to the valuation functions specified for the syntactic options. In addition, each object has methods for displaying itself and each object can be given a name by the user when it is defined.

For instance, a typical option in the syntax, DataList ::= Association of DataList, might be programmed as follows in object-oriented Pascal:

DataListAofD := object(DataList)

function Eval(InList: DataListPtr): DataListPtr; virtual;

procedure Display; virtual;

private

TheAssociation : AssociationPtr;

TheDataList : DataListPtr;

end;

function DataListAofD.Eval(InList: DataListPtr): DataListPtr;

begin

Eval := TheAssociation^.Eval(TheDataList^.Eval(nil));

end;

procedure DataListAofD.Display;

begin

TheAssociation^.Display;

write(' of ');

TheDataList^.Display;

writeln;

end;

This defines the Association of DataList option as an object that inherits from the DataList object, has data items corresponding to its constituent Association and DataList, and has methods for displaying and evaluating itself. Note how the Eval method follows the same applicative process as the valuation function from the denotational semantics:

D[[A of D]] = d. z. A[[A]] ( D[[D]] z)

First it evaluates D, the DataList, (using the hypertext database z, but ignoring the input DataList d) and then applies the evaluation of A, the Association, to this intermediate result. (The notation A[[A]] means the Association valuation of A, that is the evaluation of a given Association instance in accordance with whichever option of the Association syntax it instantiates.) The Display method similarly displays the Association using whatever method corresponds to the particular Association instance, displays the character string ' of ', and then displays the DataList with the method appropriate to its instance. This approach to polymorphic execution depending upon the particular form of the instances at runtime provides the flexibility to define methods that handle nesting of phrases. In other words, Association can take the form of any Association option and DataList can take the form of any DataList option. This allows the Hermes Language to have the kind of phrase structure that English has, with arbitrarily deep nesting of phrases.

The denotational semantics of the Hermes language is extraordinarily simple because the language has been designed to minimize the amount of programming doctrine required. In particular, there is no state change and no continuation processing because there are no assignment statements and no explicit iteration constructs. In fact, there is no store because there are no variables.[1] Moreover, there is no explicit typing. This all makes for a simple, straight-forward language structure.

Balancing the structural simplicity of the language is the wealth of individual syntax options. The quantity of options results from the history of design trade-offs detailed in Appendix B. In particular, there was a concerted effort to provide a broad range of useful functionality while restricting the possibilities for creating problematic constructs. There are over a hundred options for the Advanced Version of the Hermes Language. Most of them define permissible and useful combinations of other options and are relatively self-explanatory. A few of them require more explanation.

Informal semantics of the Hermes language.

In the following, the major options listed in the syntax above are explained briefly and examples of their use is given.

Simple DataLists. These generate ordered lists of nodes from the database. Most of the Simple options generate a list of one node. The options for DataLists signify that these lists of nodes can be generated in the following ways:

By specifying the name of a node. Names are optional for nodes. For example:

archie’s lunar habitat

By giving the Id of a node. All nodes have unique numeric identifiers that are used internally by the Hermes system.

id: 2345

By defining a Character node using any of the Character options.

substring of privacy message from 1 to 26

By defining a Number node using any of the Number options.

By defining a Boolean node using any of the Boolean options.

there are more than 3 grandchildren of sandra

All nodes in the database of a specified NodeKind

lunar habitats

All nodes in the database of a specified LanguageType

datalists that have authors that contain "Sandra"

All nodes in the database

items

that (last subject) stands for whatever the last explicitly specified subject was during evaluation. This is used in the definition of predicates, where it is not known what will be operated on by an expression

parts of inverse parts that do not equal that (last subject)

this (expression) means the item currently being operated upon. This is similar to a reference to “self” in other languages

issue_trees: issues with their this (expression)

those items refers to the last list of intermediate result items computed. This option saves intermediate results so they do not have to be recalculated in order to be displayed

prerequisite_trees: prerequisites and prerequisite _trees of those items

prerequisites and this (expression) of those items

The name of a stored DataList whose contents (not definition) is to be used

contents of archie’s problem areas

The name of a defined ResultList whose definition (not stored contents) is to be evaluated

sandra’s prerequisite problems

Computed DataLists. These generate ordered lists of nodes from the database. Most of these apply one operator to another to generate a DataList:

By combining two DataLists with a Combination

men and women

By applying an Association to an existing DataList (and traversing links to arrive at a new list).

answers of the privacy issue

By applying an Association to an existing DataList and listing the DataList with the Associations of each item listed under that item and indented. The term “with” indicates that indenting will take place for the results of the Association. Note that the Association operator is applied to the DataList results.

answers of the privacy issue with their arguments

By applying a Filter to an existing DataList (and eliminating nodes which do not pass through).

chairs that are near tables

By specifying a Graphic that is internal to a hierarchical Graphic. If the optional term “immediately” is used, then only the first level parts of the Graphic will result, and not the parts of parts, etc. as when the “immediately” keyword is absent.

chairs in (graphic) habitats

areas immediately in (graphic) habitats

By selecting a DataList as viewed within a context other than the currently active perspective

habitats in context lunar gravity

By choosing between two existing DataLists depending upon whether the first has any items when evaluated

either problem areas or approval message

By evaluating a Boolean and choosing between two existing DataLists; the second DataList is optional.

if stove is near curtains then flammability issue, else default_critic

By sorting a DataList

men and women, sorted

By removing duplicates from a DataList

men and boys, without duplicates

Simple Associations. These are computations that map or transform one list of nodes into another, based on the links coming into or out of the nodes in the first list. The transformations can take place as follows:

All links of a specified Link Type

sons

The name of a node. This is a pseudo-Association that allows internally stored names to be referenced in the Hermes language

name of sons of sandra

The id of a node. This is a pseudo-Association that allows internally stored ids to be referenced in the Hermes language

id of sandra

The creation date of a node. This is a pseudo-Association that allows internally stored names to be referenced in the Hermes language. Design environments built on Hermes could make timestamping automatic

creation date of arguments of answers of an interesting issue

The creator of a node. This is a pseudo-Association that allows internally stored names to be referenced in the Hermes language. Design environments built on Hermes could automatically stamp a newly created node with the user’s login name

creator of an interesting issue

The last modification date of a node. This is a pseudo-Association that allows internally stored names to be referenced in the Hermes language. Design environments built on Hermes could make timestamping automatic

last modification date of arguments of answers of an interesting issue

List the contexts in which something is defined. This is a pseudo-Association.

contexts that inherit from archie’s perspective

List every association

all associations of the privacy issue

Parts of a graphical hierarchy. If the optional term “immediate” is used, then only the first level parts will result, and not the parts of parts, etc. as when the “immediate” keyword is absent.

immediate parts of archie’s habitat

parts of archie’s habitat

Any Dimension (see Dimensions below). This is a pseudo-Association

length of subparts of archie’s habitat

Distance from a specified graphical object. When applied to one or more graphical objects, this returns the numeric distance(s). Distances are always measured within some implicitly or explicitly specified encompassing graphic

closest distance in feet from table in archie’s habitat

The name of a defined Association

sons

Predicates. These are Computed Associations that display their resultant lists differently in order to hide the computations that have been encapsulated in the Predicate definition. Any Association definition can optionally be stored as a Predicate. For instance, the Association

discussion: issues with their answers with their arguments

can be stored as an Association or as a Predicate. The same nodes would appear in the lists generated by the use of either, but the lists would appear differently when displayed. The results of the Association would be labeled as issues, answers, and arguments, and they would be indented accordingly to show the structure of the computation. The results of the Predicate, in contrast, would all be labeled discussion and would not be indented. To the user, it would appear that the results of the Predicate were all linked directly to the original nodes by simple links of type discussion—so the complexity of the actual computation would be hidden.

InputAssociation. This is a subclass of Associations that is used for formulating macros for input of structured data. Its definitions are identical to the corresponding Association definitions, but they are limited as to the complexity of their structure. In particular, they are limited to forms meaningful for prompting input of new nodes. Any InputAssociation can be included wherever a Association can be used, and it can be saved as an Association or a Predicate. However, the reverse is not true, and only expressions saved as InputAssociation can be used for eliciting the entry of new nodes. For instance, if discussion were saved as an InputAssociation, then the user could be prompted to enter a hierarchy of design rationale automatically. First, the system would prompt for an issue. For each issue entered, it would prompt for a series of answers. And as each answer was entered, the user would be prompted for arguments to support it.

Computed Association. These combine Simple Associations with other operations.

Apply one Association to the result of another Association

arguments of answers

Apply one Association to the result of another Association and list each item of the result of the first Association with its Associations listed under it and indented

answers with their arguments

First apply an Association and then apply a Filter

sons that are contained in siblings of sandra

Follow the in-coming Association links instead of the usual out-going links

inverse parent

Choose between two existing Association results depending upon whether the first has any items when evaluated

either sons or daughters

Evaluate a Boolean and choose between two existing Associations; the second DataList is optional.

if stove is near curtains then arguments

Combine two Associations with a Combination

sons and daughters

Select the n-th Association

the 7th son

Sort the Association results

authors of novels, sorted

Remove duplicates from the Association results

authors of novels, without duplicates

Simple Filters. Filters are conditional operators applied to each node in a DataList; if the condition is true of the node, then the node is retained (from the input list, in the output list). The following filtering operations are defined:

Check if an item is equal to (the same object as) an item specified by a DataList

habitats that have chairs that equal archie’s favorite chair

Check if an item has a name specified by a defined Character

habitats that are named the secret word

Check if an item is included in a defined DataList

chairs that are included in habitats

Check if an item includes a defined DataList

habitats that include chairs

Check if an item is of a kind defined by a NodeKind

parts of habitats that are of kind chair

Check if an item is of a type defined by a LanguageType

items that are of type distances

The name of a defined Filter

are desirable

Multimedia Filters. These include filters specific to Characters, Numbers, Booleans, Contexts, and Graphics.

Check if a substring is included that is defined by a Character

messages that include the warning string

Check if a numeric item equals an amount that is defined by a Count

that are more than 3

Check if a boolean value equals true

test conditionals that are true

Check if a specified DataList or Count of the DataList can be viewed in a Context

contexts that view more than 7 issues that include “bunk”

Check if a Context inherits from another Context

contexts that inherit from archie’s context

Check if a Context is inherited by another Context

contexts that are inherited by archie’s context

Check if a graphical item contains the graphical items that are defined by a DataList. The keyword “immediately” restricts the computation to the highest level parts.

habitats that contain chairs

Check if a graphical item is contained in the graphical items that are defined by a DataList. The keyword “immediately” restricts the computation to the highest level parts.

chairs that are immediately contained in habitats

Check if graphical items are a distance defined by a Measure from items defined by a DataList. This may optionally be quantified with a Quantifier. Distances are implicitly or explicitly measured within a Graphic

that are near most stoves in the kitchen

Check if an Attribute value equals Value

that have color is red

Check if an Attribute value equals Number

that have pen width is 5

Computed Filters. These include Filters that combine other operators.

Check if there are a certain number defined by Count of Associations. Optionally sublist the intermediate results

that have more then 3 grandchildren with those items

Check if there are a certain number defined by Quantifier of Associations that pass a Filter. Optionally sublist the intermediate results

that have all grandchildren that are of kind boy

Check a Filter only if a Boolean is true. Optionally check an alternative Filter otherwise

that if test conditional then are red, else are blue

Check if items pass both of (or one of) two Filters, depending on the Connective, logical and/or

that are more than 3 and (logical) are less than 7

Check if items do or do not pass a Filter

that are equal “bunk”

that are not named “my bunk”

that do not equal “bunk”

The semantics of the remaining media options, network options, namable terminology, and simple terminology options are straight-forward and should be clear from their BNF syntax. The media elements provide the primitive values for the content of nodes and the terminal values for the language options. However, they also include computations involving other object types. The network elements define the graph structure of the database. Some node kinds and link types have been pre-defined for internal use by the system. They are available to the user, but cannot be modified. The namable terminology elements serve mainly to provide choices for the language options. However, they can involve computations. Just like any of the above options, these can be named and saved. For example, some and several have already been defined in the Hermes seed as supplementary Quantifiers this way. It has been assumed in the examples above that several Measures have been defined, like too near or far away from. These can be redefined and personalized by users in their interpretive perspectives. In contrast, the simple terminology elements are fixed: they cannot be named or redefined and extended by users.

[1] In fact, there is a very limited state change, implemented with a specific associated store for the DataList options that (last subject), this (expression), and those items. The first two of these each require a special stack and the third requires a list to keep track of their changing values. These are simple to implement and can be easily accounted for in the denotational semantics without all the overhead of variable names and assignments.

Go to top of this page

Return to Gerry Stahl's Home Page

Send email to Gerry.Stahl@drexel.edu

This page last modified on January 05, 2004