Most languages and tagsets have personal pronouns, i.e. words like “I”, “you”, “he”, “she”, “it”, “we”, “they”. Various interrogative and relative function words (wh-words in English) are often also considered pronouns. In addition, grammars of some languages distinguish determiners while others prefer to categorize the same thing as a sort of pronouns. According to the definition of EAGLES, pronoun is a function word that replaces a noun phrase, while determiner is a function word that modifies a noun phrase. As a result, proper EAGLES-pronouns behave like nouns and determiners behave like adjectives. Note that possessive pronouns (i.e. “my”, “your”, “his”, “her”, “its”, “our”, “their”), also found in many languages, are personal possessive determiners in the sense of the EAGLES definition.

Because tagsets often disagree in what is pronoun, what is determiner etc., it is difficult to find a unifying approach. We decided to limit the number of the major parts of speech in order to minimize the cases where a word would end up with an empty part of speech. If there were a part of speech called determiner, drivers of tagsets not having determiners would either have to check whether pos = det during encoding, or they would fall back into a residual word class. On the other hand, if we tag determiners as special cases of adjectives (which is what DZ Interset does), such drivers will simply encode determiners as adjectives (provided they have adjectives—but these are much more common).

We also followed this solution with pronouns because of the following reasons:

• Although pronouns are found in most tagsets, there is much controversy about the precise extent of that category.
• Some tagsets allow for distinguishing between substantive and attributive pronouns (roughly those that behave like nouns and those that behave like adjectives). Assigning pronouns to nouns and adjectives respectively helps preserve that distinction.
• Some of the features that distinguish pronouns from real nouns and adjectives (interrogativeness, for instance) are found with adverbs and numerals as well and thus it makes sense to separate them.

The pos feature should be set to noun for real (according to the EAGLES definition) pronouns, and to adj for determiners (or attributive pronouns). Default value is noun and should be used if the correct value cannot be figured out from the source tagset (for instance, the tagset has only one tag for all pronouns and determiners; or there are words that can occur in both pronominal and determinative contexts, and the tagset does not distinguish these contexts).

The prontype feature distinguishes personal, demonstrative, interrogative, relative, indefinite, negative and other pronouns. It also distinguishes pronouns and determiners from real nouns and adjectives, from which it follows:

• It should never be empty if the word is a pronoun or a determiner. If the source tagset does not specify the subclass of the pronoun, the default value prs (personal) should be used.
• The use of prontype is not restricted to nouns and adjectives. Some adverbs and numerals are good candidates, too. However, the driver should treat this feature carefully because it is its means of restoring pronouns if they are to be encoded. The more word classes have prontype the more difficult it is to recognize real pronouns, as defined by the driver.

Other features (poss, reflex) are orthogonal to those mentioned above, although many tagsets encode them together with the main pronoun type. However, we can have relative possessive determiners (“whose”), reflexive possessive pronouns/determiners (Czech “svůj”) etc.

Our solution makes it difficult to recognize a pronoun for a target tagset that can encode it. It is especially annoying if the tagset needs to query the part of speech many times in order to select the correct features. Under such circumstances the probably best thing to do is to define a tagset-specific is_pronoun() function, e.g. as

$f{pos} =~ m/^(noun|adj)$/ && $f{prontype} ne ""