Bibliome » AlvisNLP/ML

h1. Element Expression

{{toc}}

h2. Introduction

Element Expressions is a language for exploring and querying the AlvisNLP Corpus. It can be used to test features, count elements, retrieve annotations with certain characteristics, etc. This language shares a lot of common points with XPath, so if you are, or become, familiar with XPath, then good for you. You may find the following pages also useful:

* [[Element Expression Examples]] contains several examples of expressions in increasing order of trickiness;

* [[Shell]] describes the AlvisNLP/ML Shell that you can use to train yourself to write expressions.

h2. Context Element

Expressions are evaluated within a context that includes an element. The context element can be one of the following:

* the corpus

* a document

* a section

* an annotation

* a relation

* a tuple

Some expressions are independent of the context element; their evaluation does not depend on it (for instance arithmetic operators). However the most useful expressions depend on the context element, for instance the evaluation of a feature value expression obviously depends on the context element. Wherever an expression is expected, for instance as a module parameter, the context element sould be documented.

h2. Evaluation types

An expression can be evaluated as on of four types: boolean, number, string or element list. The evaluation type should be documented along with the context element.

h3. Scalar types

h4. Boolean

The boolean type has two values: @false@ and @true@.

h4. Number

The number type is a double precision 64-bit floating point number (Java @double@).

h4. String

The string type is a 16-bit unicode character sequence (Java @String@).

h3. Element list

The element list type is an ordered collection of elements. In most cases elements in an element list are of the same type (all Annotations or all Documents etc.).

h3. Type coercion

The majority of expressions have a priviledged or primary evaluation type, however they can be evaluated into another type. The value is computed using type coercion rules:

|       |*boolean*|*number*|*string*|*list*|

|*boolean*||false=0, true=1|false="false", true="true"|empty list|

|*number* |0=false, otherwise true||decimal notation string|empty list|

|*string* |""=false, otherwise true|decimal conversion||empty list|

|*list*   |empty=false, otherwise true|element count|concatenation of static features||

Some expressions have specific coercion rules.

h2. Operator precedence and associativity

The following operators are listed in descending order of precedence. The precedence can be overriden with parentheses.

|*Operators*|*Associative*|

|@if then else@|no|

|@or@|yes|

|@and@|yes|

|@not@|no|

|@== != < > <= >= ?= ^= =^ =~ in any@|no|

|@^@|yes|

|@+ -@|yes|

|@* / %@|yes|

|unary @-@|no|

|pipe|yes|

|@.@|yes|

h2. Syntax for names

Some expressions require a name (feature key or layer name for instance). Names are single quote character sequences. The quotes can be omitted if all the following conditions are met:

* all characters are alphabetic (@A-Za-z@) or undescore (@_@)

* the name is different from any reserved word:

|@after@|

|@and@|

|@any@|

|@args@|

|@before@|

|@boolean@|

|@contents@|

|@corpus@|

|@document@|

|@documents@|

|@elements@|

|@else@|

|@end@|

|@false@|

|@if@|

|@in@|

|@inside@|

|@layer@|

|@length@|

|@not@|

|@number@|

|@or@|

|@outside@|

|@overlapping@|

|@relation@|

|@relations@|

|@section@|

|@sections@|

|@start@|

|@string@|

|@then@|

|@true@|

|@tuples@|

Note that names and keywords are case-sensitive. All keywords are all lowercase.

h2. Expression reference

In the following sections each available expression is described. The usage of the expression is given in preformatted paragraphs with the following conventions:

<pre>

  EXPR         uppercase words are variable parts of the expression usage, they are meant to be replaced either by sub-expressions, names or constants

  layer        lowercase words are keywords

  [ ... ]      parts between brackets are optional

  ( ) . + :    all other sylmbols are operators, parenthes or a column, they are part of the expression syntax

</pre>

If there is a preferred type for the expression, then this type is specified between brackets in the expression name.

h4. Boolean constant [boolean]

<pre>

  false

  true

</pre>

h4. Integer constant [number]

<pre>

  [0-9]+

</pre>

h4. String constants [string]

<pre>

  "..."

</pre>

String constants are double quoted character sequences. The usual Java escape sequences apply.

h4. Boolean operators [boolean]

<pre>

  LEFT and RIGHT

  LEFT or RIGHT

  not EXPR

</pre>

@LEFT@, @RIGHT@ and @EXPR@ are evaluated as booleans with the same context element.

Binary boolean operator evaluation is short-circuited.

h4. General comparison [boolean]

<pre>

  LEFT == RIGHT

  LEFT != RIGHT

</pre>

@LEFT@ and @RIGHT@ are evaluated as the same type with the same context element. If @LEFT@ is an expression of scalar type, then its type is used. Otherwise, the type of @RIGHT@ is used.

h4. Integer comparison [boolean]

<pre>

  LEFT < RIGHT

  LEFT > RIGHT

  LEFT <= RIGHT

  LEFT >= RIGHT

</pre>

@LEFT@ and @RIGHT@ are evaluated as numbers using the same context element.

h4. String comparison [boolean]

<pre>

  LEFT ?= RIGHT

  LEFT ^= RIGHT

  LEFT =^ RIGHT

</pre>

@LEFT@ and @RIGHT@ are evaluated as strings using the same context element.

Meaning of operators:

|@?=@|contains|

|@^=@|starts with|

|@=^@|ends with|

h4. String concatenation [string]

<pre>

  LEFT ^ RIGHT

</pre>

@LEFT@ and @RIGHT@ are evaluated as strings with the same context element.

h4. Regexp match

<pre>

  TARGET =~ "PATTERN"

</pre>

@TARGET@ is evaluated as a string with the same context element. @"PATTERN"@ is a string constant containing a regular expression in "Java syntax":http://download.oracle.com/javase/7/docs/api/java/util/regex/Pattern.html

If evaluated as a boolean, then this expression returns either the target matches the pattern.

If evaluated as a number, then this expression returns the number of non-overlapping matches of the pattern in the target.

If evaluated as a string, then this expression returns the first match of the pattern in the target.

If evaluated as an element list, then this expression returns an empty list.

h4. Arithmetic [number]

<pre>

  LEFT + RIGHT

  LEFT - RIGHT

  LEFT * RIGHT

  LEFT / RIGHT

  LEFT % RIGHT

</pre>

@LEFT@ and @RIGHT@ are both evaluated as numbers with the same context element.

h4. Unary minus [number]

<pre>

  - EXPR

</pre>

@EXPR@ is evaluated as a number with the same context element.

h4. Dictionary lookup [boolean]

<pre>

  EXPR in "FILE"[:"ENCODING"]

</pre>

@EXPR@ is evaluated as a string with the same context element. @"FILE"@ is a string constant containg the path to a dictionary file. @"ENCODING"@ is a string constant containing the name of the dictionary file character set. If the encoding is omitted, UTF-8 is assumed.

The dictionary file must contain one entry per line. This expression returns true if and only if the dictionary contains the first operand.

h4. Feature

<pre>

  KEY

</pre>

@KEY@ is a name. This expression returns the last value of the feature with key @KEY@ of context element.

If this expression is evaluated as a boolean, then it returns @true@ if and only if the context element has a feature with key @KEY@, +even if the feature value is an empty string+.

h4. Any feature value equals

<pre>

  any @KEY@ == @EXPR@

</pre>

@KEY@ is a name. @EXPR@ is evaluated as a string with the same context element.

This expression returns true if at least one of the values of the feature with key @KEY@ in the context element equals @EXPR@.

h4. Annotation positions [number]

<pre>

  start

  end

</pre>

These expressions return the start or end position if the context element is an annotation. Otherwise it returns @0@.

h4. Element length [number]

<pre>

  length

</pre>

If the context element is an annotation, then this expression returns its length.

If the context element is a section, then this expression returns returns the length of the section's contents.

h4. Reference

<pre>

  $REF

</pre>

@REF@ is a name. This expression returns the value of the named reference.

Available references should be documented by the module.

h4. Section contents [string]

<pre>

  contents

</pre>

If the context element is a section, then this expression returns its contents. Otherwise the empty string is returned.

h4. Conditional

<pre>

  if CONDITION then TRUE else TRUE

</pre>

@CONDITION@ is evaluated as a boolean with the same context element. If the result is, then @TRUE@ is evaluated as the same type with the same context element. Otherwise @FALSE@ is evaluated as the same type with the same context element.

h4. Union [list]

<pre>

  LEFT | RIGHT

</pre>

@LEFT@ and @RIGHT@ are evaluated as element lists with the same context element. This expression returns the concatenation of the two results.

Elements in the result list are not reordered. Duplicate lements remain.

h4. Path

<pre>

  LEFT . RIGHT

</pre>

@LEFT@ is evaluated as an element list with the same context element, then each element of the result is used as the context element to evaluate @RIGHT@.

If this expression is evaluated as a boolean, then it retuns @true@ if any evaluation of @RIGHT@ as a boolean is true.

If this expression is evaluated as a number, then it returns the sum of all successive evaluations of @RIGHT@ as a number.

If this expression is evaluated as a string, then it returns the concatenation of all successive evaluations of @RIGHT@ as a string.

If this expression is evaluated as a list, then it returns the concatenation of all successive evaluations of @RIGHT@ as a list.

h3. Element navigation expressions [list]

Element navigation expressions returns elements according to a navigation specification. The following subsections describe each available specification.

A specification can be followed by filters and ranges. The order of filter and ranges specifies the order in which they are applied. If a range follows a filter, then range is applied after the filter.

h4. Filters

<pre>

  SPEC ( EXPR )

</pre>

@SPEC@ is a navigation specification. @EXPR@ is evaluated as a boolean with the current element as the context element.

The expression returns the list of elements for which @EXPR@ was evaluated as @true@.

h4. Ranges

<pre>

  SPEC ( N )

  SPEC ( N : M)

  SPEC ( : M )

  SPEC ( N : )

</pre>

@SPEC@ is a navigation specification. @N@ and @M@ are integer constants.

The returned list is a sublist of the list returned by @SPEC@:

|@N@|a singleton list with the @N@th element|

|@N : M@|the sublist from the @N@th (inclusive) to the @M@th (exclusive) elements|

|@: M@|the sublist from the start to the @M@th element (exlusive)|

|@N :@|the sublist from the @N@th element (inclusive) to the end|

List indexes are zero-based: @0@ is the first, @1@ is the second, etc. If @N@ or @M@ are negative, then the length of the list + 1 is added to their value: @-1@ is the last (inclusive).

If the indexes are out of the list boundaries then the index is "cropped".

h4. Element corpus

<pre>

  corpus

</pre>

This expression returns the currently annotated corpus.

h4. Element document

<pre>

  document

</pre>

This expression returns a singleton list containing the document to which the context element belongs.

If the context element is a document, then this document is returned.

If the context element is the corpus, then the empty list is returned.

h4. Element section

This expression returns a singleton list containing the section to which the context element belongs.

If the context element is a section, then this section is returned.

If the context element is the corpus or a document, then the empty list is returned.

h4. Tuple relation

<pre>

  relation

</pre>

If the context element is a tuple, then this element returns a singleton list with the relation to which the tuple belongs. Otherwise it returns the empty list.

h4. Corpus documents

<pre>

  documents [ : ID ]

</pre>

@ID@ is a name. If the context element is the corpus, then this expression returns a singleton list containing the document with the identifier @ID@. If @ID@ is omitted, then this expression returns a list containing all documents in the corpus.

If the context element is not the corpus, or there is no document in the corpus, or there is no document with the specified identifier, then this expression returns the empty list.

h4. Document sections

<pre>

  sections [ : NAME ]

</pre>

@NAME@ is a name. If the context element is a document, then this expression returns a list containing all sections in the document with the name @NAME@. If @NAME@ is omitted, then this expression returns all sections of the document.

If the context element is not a document, or there is no section in the document, or there is no section with the specified name, then this expression returns the empty list.

h4. Section annotations

<pre>

  layer [ : NAME ]

</pre>

@NAME@ is a name. If the context element is a section, then this expression returns a list containing all annotations in the layer named @NAME@. If @NAME@ is omitted, then this expression returns all annotations of all layers of the section.

If the context element is not a section, or there is no layer with the specified name, or there is no annotation in the section, or the layer with the specified name is empty, then this expression returns the empty list.

In all cases, the list of annotations is sorted by standard order (increasing start, then decreasing end) and duplicates are removed.

h4. Section relations

<pre>

  relations [ : NAME ]

</pre>

@NAME@ is a name. If the context element is a section, then this expression returns a singleton list containing the relation in the section with the name @NAME@. If @NAME@ is omitted, then this expression returns all relations of the section.

If the context element is not a section, or there is no relation in the section, or there is no relation with the specified name, then this expression returns the empty list.

h4. Relation tuples

<pre>

  tuples

</pre>

If the context element is a relation, then this expression returns a list of all tuples of the relation. Otherwise it returns the empty list.

h4. Tuple arguments

<pre>

  args [ : ROLE ]

</pre>

@ROLE@ is a name. If the context element is a tuple, then this expression returns a singleton list containing the annotation which is the argument of the tuple with the role @ROLE@. If @ROLE@ is omitted, then this expression returns a list containg all arguments of the tuple (in no particular order).

If the context element is not a tuple, or the tuple has no arguments, or if the tuple does not have an argument with the specified role, then this expression returns the empty list.

h4. Reverse tuple lookup

<pre>

  tuples : RELATION [ : ROLE ]

</pre>

@RELATION@ and @ROLE@ are names. If the context element is an annotation, then this expression retuerns a list containing all tuples that satisfy all the following conditions:

# the tuple pertain to the relation with name @RELATION@ in the same section

# the annotation is the argument of the tuple with role @ROLE@, if @ROLE@ is omitted, then the annotation is an argument of the tuple regardless of the role

If the context element is not an annotation, or the section does not contain a relation with the specified name, then this expression returns the empty list.

h4. Annotation siblings

<pre>

  after : NAME

  before : NAME

  inside : NAME

  outside : NAME

  overlapping : NAME

</pre>

If the context element is an annotation, then this expression returns a list of annotations in the layer with name @NAME@ in the same section. The annotations included in the result list depend on the keyword:

|@after@|start after the context annotation end|

|@before@|end before the context annotation start|

|@inside@|fully included in the context annotation span|

|@outside@|fully includes the context annotation|

|@overlapping@|overlaps (broad sense) the context annotation|

If the context element is not an annotation, or the section does not have a layer with the specified name, then this expression returns the empty list.

In all cases, the returned list is sorted in standard order.