1. Introduction

a. The concept of the 'word' as the basic unit of linguistic analysis is central to many grammatical and lexicographical resources. Though most language users possess an intuitive notion of the word, its definition and analysis has been an enduring problem for linguists and general linguistics. Recent studies making use of the concept of collocation have demonstrated that distinct meanings are often communicated in fixed phrases or combinations of words, quite distinct from the semantic analysis of individual words in the phrase. Morphological analysis of individual words can also be ambiguous, as in the case of the article tw'n which can be identified as genitive plural, but its gender cannot be decided at the word level. This ambiguity is removed when the word is analyzed as part of a word group, e.g. tw'n lovgwn aujtou'.

b. A further justification for considering the word group as the basic unit of analysis is that relations between the members of the group, e.g. a head-term with a genitive modifier take place before the clause level. This allows the clause level analysis to focus on the ordering and interaction of its components (e.g. subject, predicate, complement and adjunct [see Clause Level Annotation]). Thus many features that have traditionally been treated under the label of syntax can be dealt with at the word group level.

2. Definitions

3. Features analyzed at the word group level

a. Though not formally recognized in the current elements and attributes, it is helpful to divide the features analyzed at the word group level according to whether they belong to the field, tenor or mode of discourse.

3.1. Semantic domains and relations (Field)

a. Each word within the group is marked with its major domain number(s) from the Louw-Nida dictionary. If a word is classified in more than one major domain each domain is noted in a comma separated list in the order they appear in the index of Louw-Nida, e.g. 88, 65, 57.

b. The semantic domain of a word group is taken to be that of its head term. If there are more than one possible domain for this term, the annotator must select just one from this list. The entry for a word in the Louw-Nida lexicon often includes citations associated with a specific domain classification. These should be consulted when selecting the domain classification of a head term. In addition, the annotator should pay attention to the domains of surrounding words, as by collocation they may help the disambiguation process. [[Do we need to specify criteria for this selection? What are they?] (-)]

c. Each word and word group within the discourse is assigned a unique identifier. For each word group the identifier of its head term is marked. For the other words within the group, each one is marked with an indication of which other word within the group it modifies and an indication of the kind of modification relationship (see Definitions).

d. A modifier can only modify one other word within its group. Thus only the closest semantic relation is indicated. This explains why the preposition katav in the group to; kat« ejme; provqumon is said to modify ejmev and not provqumon. In cases where a modifier appears to modify more than one word, for instance the article in Phlm. 2 tw'/ ajgathtw'/ kai suvergw'/ hJmw'n, it should be marked as modifying the word closest to it. So, tw'/ is annotated as a specifier of ajgathtw'/ (Higher level annotation can mark the fact that is serves as a modifier for two or more words).

3.2. Participant reference type (Tenor)

b. This annotation can be added in-line within the same document or out-of-line in the document or in an external document through links referencing word identifiers. The latter option is recommended as it allows a word group node to have only words as direct children. Examples of both in-line and out-of-line participant annotation are provided in section 5.

d. This proposal makes no attempt to provide criteria for making decisions on whether such phrases should be treated as a single reference or two distinct references.

3.3. Boundaries, punctuation, lexical form, and morphological features (Mode)

a. Each word and word group in a discourse is given a unique identifier allowing reference from other elements and documents. [[Should we specify these identifiers as globally unique to the OpenText.org document universe? Or is the document URL+local uniqueness enough?] (-)]

b. Punctuation and any structural and orthographic markings should be marked in-line at the point at which they occur. If the mark occurs between word groups or at the end of a group it should be marked after the end of the group marking (i.e. it should not be included in the word group). For instance: [Ma'rko"] {,} [«Arivstarco"] {,} [Dhma'"] {,} [Louka'"] {,} [oiJ sunergoiv mou] {.} (word group boundaries = [], punctuation boundaries = {}).

d. Conjunctive particles that do not function within the group (see 3.1) but at a higher level of discourse should not be included in the scope of a word group if this is possible. Thus [Pau'lo"] kai; [Timovqeo"] and not [Pau'lo"] [kai; Timovqeo"]. The annotation of these elements is discussed in the clause level specification (see Clause Level Annotation).

e. This separation is not always possible, as in the case of post-positive conjunctive particles, e.g. [cara;n ga;r pollhvn]. However, it might be justified to break the word group, allowing for the intervening element, e.g. [cara;n]^wg1a ga;r [pollhvn]^wg1b. This is not two separate word groups, but one (wg1) in two parts (a & b) with an intervening word.

f. Variation in word order within a clause may also present cases where a word group is split by another word group, e.g. [me]^wg11a [e[cei"]^wg12 [koinwnovn]^wg11b. Here word group wg12 separates me (the head term of group wg11) from its definer koinwnovn. N.B. Future versions of this specification will specify how these instances should be handled more clearly.

g. The lexical form of each word element is marked as an attribute of the element containing the actual (inflected) form of the text. Verbs should be given in the first person singular present active indicative form (-w form), unless an active form is not found in Hellenistic Greek, then the middle should be used (-omai form). [[Should we specify a standard format for lexical forms? e.g. -w forms of verbs always (even for so-called deponents) etc.] (-)]

4. Elements and attributes available for word group annotation

4.1. Namespace

a. The namespace for elements at the word group level is http://www.OpenText.org/ns/word-group.

4.2. Main elements

a. The word group namespace is declared as default on the group element and is therefore applied to all child elements without a prefix (this is a FIXED attribute in the group declaration). The group element has a prefix referring to the same namespace, declared on the document element (e.g. xmlns:wg="http://www.OpenText.org/ns/word-group"). This arrangement is selected to overcome the current lack of support for namespaces in XML DTDs (the group element is actually defined as wg:group in the DTD; see section 7).

b. The element to mark a participant reference (<wg:part>) also carries the word group namespace prefix. Used in-line it surrounds <w> elements within a <wg:group> element (e.g. <wg:part> <w> oJ </w> <w> pisteuvwn </w> <w> ejn </w> <w> qew'/ </w> </wg:part>). Used out-of-line outside of the co-textual data or in an external document the element is an extended-link with <start> and <end> locators (see 4.7).

4.3. <wg:group> element

4.4. <word> element

4.5. <wg:part> element

5. Examples

6. Document Type Definition