In the European grammaticographic tradition, some of the categories of grammatical words, for instance pronouns, have been recognized since ancient Greek and Latin grammar. However, a major distinction between lexical words and grammatical words only came up with linguistics as a discipline, thus, at the beginning of the 19th century. In traditional Chinese grammaticography, a distinction between ‘real words’, i.e. content words (shící, 实词), and ‘unreal’ or ‘empty words’, i.e. function words (xūcí, 虚词), was made from the beginning. In both of these traditions, the distinction between the two main classes was categorical, i.e. there was no transition zone for borderline cases.
The idea that grammatical formatives originate in lexical items can be traced back at least to the 18th century (Lehmann 2015, §1). A set of standard examples from the Latin-Romance history is provided in Schlegel 1818. These include the development of articles and of auxiliaries in the Romance languages. Schlegel is the first to draw attention to the grammaticalization of Latin habere ‘have’ and tenere ‘hold’ in Ibero-Romance and to the factors of desemanticization and expansion observable in such cases (op.cit. 29). The idea that grammatical formatives stem from lexical items and that affixal morphology evolves by coalescence – called “agglutination” at the time – of lexical words with grammatical formatives was current in 19th century European linguistics. It plays an important role in the work of Franz Bopp (Lehmann 2016) and was the basis of Wilhelm von Humboldt's evolutive typology, taken up in Gabelentz 1891.
The first linguist to systematize some empirical evidence and to propose the concept of grammaticalization is Antoine Meillet (1912). According to his often-cited definition, grammaticalization is
l'attribution du caractère grammatical à un mot jadis autonome
(Meillet 1912: 131)
(“the attribution of a grammatical character to a word that was once independent”). Meillet also insists on the parallelism in the “weakening” of the phonological form and of the meaning.
The concept and term of grammaticalization and the underlying factor of expansion are taken up and developed in Kuryłowicz 1965. The last publication in the prehistory of research into grammaticalization is Givón 1971, whose dictum “today's morphology is yesterday's syntax” developed into a slogan. Since then, grammaticalization has become an established research field in linguistics. A set of standard works and of collective volumes has appeared, several of which are listed in the bibliography. There was also a series of four conferences called ‘New Reflections on Grammaticalization’. The role of grammaticalization in cognitive linguistics is discussed in van der Auwera et al. 2019.
In retrospect, it is to be noted that the Romance languages have played a major role in the history of research in grammaticalization, from the beginnings with Schlegel to modern treatments in Hennemann & Meisnitzer (eds.) 2022. This is no coincidence. On the one hand, the Romance languages, including Latin, have a documented history of two and a half millennia which is full of grammaticalization phenomena. On the other hand, both Latin and the Romance languages have been prominent in occidental education, so that all European scholars of linguistics from Schlegel up to the end of the twentieth century knew Latin and had some knowledge of Romance languages. The data are there and have only to be used. Non-Romance and other Non-Indo-European languages have been benefitting from grammaticalization research increasingly since Givón 1971.
Reference
Hennemann, Anja & Meisnitzer, Benjamin (eda.) 2022, Contact-induced and cognitively motivated grammaticalization and lexicalization processes in Romance languages. Heidelberg: Winter (Studia Romanica, 232)
van der Auwera, Johan & Van Olmen, Daniël & Du Mon, Denies 2019, “Chapter 11: Grammaticalization”. Dąbrowska, Ewa & Divjak, Dagmar (eds.), Cognitive linguistics. Key topics. : De Gruyter Mouton (Mouton Reader); 212-230. 
]