The model suggested by Pullum (1994) in order to define the word classes (i.e. the traditional ‘parts-of-speech’ [PoS]) seems to be the most adequate among the many models that have been proposed. A word class is a set of linguistic objects having common traits which, on their turn, are implemented by particular values (see fn. 1 for the graphic devices to distinguish between ‘categories’, ‘features’, and ‘values’). Thus, verbs may share with nouns and pronouns features such as gender and number. For instance, a possible implementation of the trait – technically called ‘feature’ – mood can be subjunctive while a possible implementation of the feature gender is feminine. Some values may be shared by different categories or features – though not all at the same time. Word classes are not water-proof boxes and lexemes may leak from a class to another one. For instance, the -ing forms show different functions according to different contexts (for a ‘construction grammar’ approach see Croft, 2007: 421, who speaks of «overlapping categories of formatives which represent their diverse distributional behaviour»): These cars want washingVBcarefully versus These cars want careful washingN. Category shifts (i.e. transcategorizations) are caused by the morphosyntactic constructs they appear in. Engl. bar is properly the imperative of the verb to bar “to prevent”, but in a sentence like Everyone is leaving bar Ernst, it has prepositional function.