This book – presented as a Ph. D. thesis at the University of Uppsala, Sweden – comprises analyses of eleven frequently occurring epistemic expressions in present-day spoken Estonian. The constitutive elements in these expressions are finite verb forms in any of the three persons – for example, ma arvan ‘I guess’, ütleme ‘let's say’, kule ‘listen!’, on ju ‘(it) is surely’. In addition to the verb form, which is either in the indicative or imperative mood, there may be another element, typically a personal pronoun or verbal suffix (1st or 2nd person), and even a clitic particle (on+ ju). Semantically, most of the verbs denote human cognition or speaking. An exception is the verb olla ‘to be’ (cf. on ju above).

The aim of the study is to show how these epistemic expressions have become or are in the process of becoming grammaticalized (de facto, often lexicalized) into fixed phrases, which the author alternatingly calls “particles,” “particle like expressions,” or “adverbs.” The phenomenon is not unknown in other languages (see, e.g., Kärkkäinen 2003 on I think, or Östman's classical You know, 1981). Some of the Estonian expressions that Keevallik has studied have previously been looked into from a different theoretical perspective, such as cognitive linguistics. Keevallik's work is a first and at the same time an ambitious survey covering a vast area of conversational practices.

The data used in this study predominantly come from the author's own recordings, 324 phone calls altogether. In addition, she has had access to the spoken language corpus at the University of Tartu, Estonia. There are both private phone calls, and business (i.e. sales) calls, and the number of speakers is over 300. This will solve the problem of idiosyncrasy, but not quite the issue of genre specificity. The author has examined an impressive amount of nearly eleven hours of spoken data.

From the recordings Keevallik has isolated more than 2,000 examples of the items in question. In her database, some of them appear more than 350 times, while the two least frequent ones – ütleme ‘we say’ and olgu ‘let (it) be’ – have 49 and 58 occurrences, respectively. The criterion for deciding what to include in the study was the number of occurrences, so frequency was decisive for regarding the item as having become routinized. While the criterion is recommendable in itself, one would have expected more discussion of the great differences between the numbers of items under scrutiny, as well as the proportion of occurrences that still passed as “literal.” There are expressions that are lexicalized beyond doubt – like palun ‘excuse me’ (lit., ‘I ask’) – and that are hardly ever used in the literal sense in a database of this size, while others are still evidently in the process of becoming fixed phrases. Also, the frequencies obtained could have been compared with those of such expressions that are unquestionable particles, like ju ‘you know’ or ega (not translatable).

The book is divided into four chapters, two of which are introductory, presenting the theoretical framework and the data. The theoretical chapter on previous work is well written and informative; however, more discussion of issues between grammaticalization and lexicalization would have been in order. The third chapter, “The analyses,” includes all the analyses of the eleven items, in alphabetical order. In my opinion, many of the treatments would have deserved chapters of their own. The final chapter is a theoretically oriented summary of the generalizations that arise from the individual analyses.

The analytical sections have the following format: as a title, each of them has been given the name of one or two actions, presumably representing the most prototypical environments of the element in the database, such as “Focusing and explaining with vaata/vat ‘look’” or “Expanding, resuming and repairing with tähendab ‘(it) means’.” Subsequently, the presentation of each item proceeds from examples of what the author calls the “literal use” to the more grammaticalized ones. The line between literal and non-literal use is, of course, a hard one to draw. There is a lot of useful distributional information on each item: whether it can form a turn of its own, and whether it can appear turn initially, internally, and/or turn-finally. However, it remains an open question in what ways these different positions are tied to different interactional practices. The longest treatment, about 30 pages, is given to tähendab, whereas only 4 pages are devoted to the epistemic expression ma arvan ‘I think’. Should this be taken as an indication of the respective distances these items have proceeded on the grammaticalization path?

The main methodological approach used in the analysis is Conversation Analysis (CA), which presupposes a detailed analysis of the interactional situation, paying attention to the speakers' joint achievement in reaching understanding. This strictly empirical method resembles in many ways the so-called discovery procedure of linguistic structuralism as practiced in the 1950s and 1960s: The researcher is not supposed to appeal to the speakers' intentions, desires, or other psychological states in explaining the distribution of the forms, but all the information should be “there,” observable on the surface. A major difference is, however, that CA emphasizes that the researcher should possess member's knowledge of the culture he or she is working with – so that, in an important sense, the analyst would be able to detect and understand the orientations of the speakers, however faintly they may be hinted at on the surface of the talk. Owing to the very extensive database used in Keevallik's work, quantitative information tends to dominate here over a detailed analysis of speakers' orientations.

In addition to CA (and interactional linguistics), the author makes use of some ideas of the theory of grammaticalization. Within this framework, it is mainly certain historical linguists who have looked into the typical ways that constructions emerge from less fixed expressions, or grammatical elements develop from lexical units. From another perspective, a growing number of pragmaticians and interactional linguists are working with synchronic data – that is, conversational material – and trying to extract chains of grammaticalization from variation. It is in this vein of scholarship that Leelo Keevallik's work belongs.

What makes the study of “particles” and fixed phrases so exciting is that they do not belong to the nucleus of syntax, nor do they undergo basic syntactic processes: They do not take part in the constituent structure of a sentence, and semantically they do not convey anything to the propositional content of a sentence. In other words, as far as traditional grammar is concerned, they are regarded as peripheral. However, the items play an important – often crucial – role in the constitution of utterances and turns, sometimes even “larger conversation structural units,” the chunks of speech through which participants in a conversation carry out their actions and manage interaction cooperatively. Works like Keevallik's book are the beginning of building a bridge between historical linguistics and conversational studies. The work could be complemented with evidence from other genres such as dialogue in older literature and plays.

With respect to grammaticalization, Keevallik's work can be seen as a contribution to the line of thinking that there may be – at least for some items – a complex set of alternative grammaticalization paths. Her analysis is not totally without problems here, though. The postulating of diverging paths should, in my opinion, require more substantial support from specific empirical evidence. In this book, the reader must be content with cryptic claims of possible schemata, like the one for kuule ‘listen’ (73):

Be that as it may, the picture emerging from natural conversational data is messier and much more complex than textbook examples of grammaticalization. The picture is particularly messy because there is much variation in the cognate forms of the “same” lexical-grammatical item (e.g., oota ∼ ota ∼ ot ∼ oot ‘wait’). The relevance of phonological variation to the issue of grammaticalization would have deserved more attention. It is not necessarily a problem: In spoken Estonian, one easily finds instances in which any word gets abbreviated when quickly spoken, as in hakasin ∼ aksin ‘I began’, or sellepärast ∼ sellepst ‘because of’. But I think it would be good to know more about the distribution of the respective particle variants, because the phonological truncation of forms is one of the criteria on which decisions are made as to whether an item has become grammaticalized.

Despite my critical remarks, I find Keevallik's book an inspiring and provocative piece of academic work. The results would have been more robust had the author restricted her analyses to fewer items and inspected them with more care, especially from an interactional perspective. However, as it stands, it will open up many new lines of study, and it also makes us want to look for similar processes in the languages we study.