In CLDF wordlists are marked by a common property dc:conformsTo with value http://cldf.clld.org/v1.0/terms.rdf#Wordlist on a TableGroup.

In the simplest case, a wordlist is just a list of triples (language, concept, word), thus in CLDF only a FormTable is required.

Providing support for the handling of lexical data is one of the core goals of CLDF. By now, many examples on how to handle certain kinds of data have been presented in talks but also implemented in different tools, including LingPy and EDICTOR. Lexical data is to a large extent much more amenable to strict cross-linguistic formalization than other types of linguistic data, but it is obvious that not all language-specific problems can be handled at once. A general idea behind CLDF is to not be intimidated by apparent complexity but instead to try to amplify the power of standards step by step.

The framework for lexical specifications, appropriate for word list data, is roughly oriented toward the model of the linguistic sign (which can be a word but also a morpheme) which consists of a language in which the sign is expressed, a meaning that the sign expresses, and a form that denotates the meaning.

• Language specification should ideally link to reference catalogs like Glottolog, but specifications of linguistic varieties may need to be more fine-grained than e.g. Glottolog allows at the moment.
• Meaning specifications again should use reference catalogs like Concepticon. This may not be possible, though, in very specific dialect datasets where many cultural terms are of interest when studying language variation.

It should be noted that both, Glottolog and Concepticon, offer ways to add new language varieties or concepts via contributions to the respective database.

Handling of the linguistic form is more complex and requires different layers of detail, starting from rough orthographic representations, up to sound sequence specifications which are cross-linguistically comparable. While efforts to address this problem are under way (see for example the specification of orthography profiles in the Unicode Cookbook for Linguists), no commonly used standard has evolved so far.

In addition to the trias of language, form, and meaning, the word list specification of CLDF also handles relations between the entities. Relations come in two general flavors: relations across languages, and relations within one language. The former are reflected in detailed specifications for the handling of different kinds of cognacy in a broader sense (akin to the concept of homology in biology, thus including borrowings, List 2016). Language-internal relations point to language-internal cognacy, that is, form-associations within one language across different meanings, ranging from simple cases of colexifications up to fine-grained accounts on internal reconstruction. Language-internal relations require additional markup that indicates relations of forms within one language using different kinds of specification.

See FormTable.

Cognate annotation is usually handled by grouping word forms in a given word list into clusters. Words in one cluster are usually supposed to be cognate in the one or the other sense. Apart from this coarse cognate relation, however, examples for more fine-grained analyses are available, and will be supported by cldf. Often, word lists use the same identifiers for unrelated words, pre-supposing that cognacy is only defined within one concept. This is the current practice in the Global Lexicostatistical Database, but given that cognacy may easily also be defined across meanings, cldf favors a strict coding in which the implicit assumption of not annotating cognates within more than one concept is made explicit by using different identifiers. The conversion is usually trivial, but it should be kept in mind when preparing datasets, and our cldf checks test automatically, whether a dataset annotates cognates across concepts (cross-semantic cognates). In addition to abstract cognate relations, distinct relations between sound segments, the supposed cognacy of sounds, are handled in alignments, which are added as a separate column and represent the homology of sounds across languages within one cognate set by adding gap-characters (a simple dash: "-") in those positions where a sound in one language corresponds to no sound in another language. An important addition to the currently common annotation of full cognacy are partial cognates. Partial cognates require a morphological segmentation of the respective Segments in the form representation, and allow to define which parts of a common word form are cognate across languages.

All in all this, gives us the following additional columns currently supported by cldf:

A note on protoforms: So far, we have not yet decided how to handle proto-forms, and adding proto-forms in an extra column would be quite redundant. A workaround would be to allow protoforms empty, so that it would be enough to add one protoform. Another possibility would be to include the proto-language as a separate language.

General Suggestions: If we allow for annotations for cluster data that the relevant information is stored in a single cell of any of the relevant rows in the table, we could easily handle cases like proto-forms without being forced to add this information in a separate row that would anyway be incomplete. Computationally, this would be feasible, as it would require at most a second parse of the data after pre-parsing them initially. Even simpler, in tools like LingPy and EDICTOR, this parse could be done dynamically, if the information is needed.

An important and new aspect of annotation in cldf which is implemented by EDICTOR and analysable within LingPy are language-internal cognate sets. The basic idea is to add a new column specifying the conceptualization of a word form. This is important in derived word forms and compounds. For example, the compound sunflower is effectively built from two words, sun and flower. If morphemes are segmented in the Segments of a word list, one can add a new column called Morphemes in order to specify the conceptualization. This is a space-segmented string in which each part annotates the original meaning of one of the morphemes. For sunflower it would just be sun flower. When comparing inside the English words of a given word list, one would then be able to retrieve that the word sunflower is partially cognate with sun and with flower. This also allows for an automatic annotation of cross-semantic relations across languages, since, if one language colexifies two concepts, like, e.g., "arm" and "hand" one could annotate this as arm-hand (one morpheme) in both semantic slots, and compare both the "arm" and the "hand" slot with other languages for cognacy. In this way, one can also handle dependencies in word lists resulting from lexicostatistical data encoding which separates cognate sets due to their meaning.