The dialects grouped under the name Alemannic span four different countries: Alemannic is spoken in Southwest Germany, in the northern part of Switzerland, in Vorarlberg, a region in Austria, and in Alsace in France. Alemannic is divided into several sub-dialects. Map (1a) shows the widely accepted division according to Wiesinger (1983).5 There is a well-known distinction in the vowel system between Swabian and the other Alemannic dialects: Swabian shifts certain long vowels into dipththongs (like Standard German) whereas the other Alemannic dialects kept the older stage of the monophthong.6 The following two7 Maps 1a and 1b show the distribution of the diphthong /au/ vs. the monopththong /uu/ in the lexical item braun (=brown), as they were produced by the speakers in a translation task (SynAlm FB1_6-2):

As can be seen, the distribution of the monophthongs follows rather strictly the established isoglosses and thus shows that – at least on the phonological level – the isoglosses are still intact. Map 1b uses the same geographical coordinates but instead of marking the dialectal borders, the political ones are displayed. The region south to Freiburg im Breisgau is the one where the dialectal border cross-cuts the political border between Germany and Switzerland. Thus, the isogloss between Swabian and Middle and High Alemannic is not constituted by the River Rhine.

Some more fine-grained sub-divisions within the Alemannic region could be reproduced with data from SynAlm as well, as Map 2a and 2b show. On these maps the results for different versions of the particle ge – used in motion verb constructions are displayed (see the next sub-section for detailed data and discussion). This particle is realized with different vowels in the various sub-dialects, therefore labeled with gV in the following. The data displayed on these maps stem from a translation task and since this particle does not occur in Standard German, any type of interference effect can be excluded.

Both maps show the stability of the dialectal areas and how they cross-cut the political border. In Map 2a and 2b, the blue dots (=go) appear in the High Alemannic area which is spoken in Germany and Switzerland. These two examples from phonology were meant to show for now that (i) the isoglosses are indeed stable, and (ii) that the data gained by SynAlm are reliable.

The project SynAlm ran for five years during which seven questionnaires were sent out. The number of informants could not be kept constant for the whole period. From about 1000 from the first round, 517 sent back the last questionnaire. Fortunately, these informants covered nevertheless the whole area, but the number of informants per measuring point decreased. The dots in the maps represent the 350 measuring points. As the number of informants per point is not constant, a uniform colouring can also mean that there is only one informant per point. The project itself focussed on (contrastive) fine grained grammaticality judgments. But due to a range of translation tasks, phonological and lexical data were also gained. Most of the data to be discussed in the following8 stem from judgment tasks.

In this section, I will show that the traditional Alemannic borders, which are based on phonological phenomena, are also relevant for morpho-syntactic patterns. For this, we can consider again the particle gV that typically occurs with motion verbs and has already gained some attention in the literature on Swiss German, e.g. Lötscher (1993), Schönenberger & Penner (1995) or Riemsdijk (2002). An example from Swiss German is given in (1):

There is not only variation with respect to the form, as illustrated with Map 2 – we find in addition syntactic differences, as discussed in Brandner & Salzmann (2009; 2012). One of the most prominent differences is the relative position of the particle within the infinitival complement. Consider (1) where the particle precedes the arguments of the infinitive – in contrast to (2):

This difference in position goes hand in hand with differing restructuring possibilities while the interpretational properties as a single event are kept constant. To account for this situation, Brandner & Salzmann (2012) propose two different grammaticalization scenarios for the two languages. They assume with Lötscher (1993) that the origin of the particle gV is a shortened version of the allative preposition gegen (=against, towards) which then came out as gen (=towards), resulting due to word-final n-drop in Alemannic in gV. As can be seen in Map 2, the sub-dialects of Alemannic differ in the vowel. Due to the low vowel in Swiss-ALM, the preposition is reanalyzed as the verb gehen (to go) which is homophonous in its infinitival form with the particle (go/ga). Exactly because of this similarity, the construction is often referred to as “verb doubling” in the literature, see e.g. Schönenberger & Penner (1995) and Riemsdijk (2002). The particle thus occupies in fact a verbal head within the VP, with a syntax akin to those of modal verbs.10 This analysis is corroborated by the fact that in Swiss-ALM this kind of verbal doubling has spread to other verbs, e.g. choo (come), lo/lasse (let) and aafange (start).

This reanalysis did not happen in German-ALM. The infinitival form of gehen is the same as in Swiss-ALM, go/ga. As this form does not match with the vowel of gi/ge, a relevant pre-condition for the reanalysis process is not met. Consequently, ge/gi has kept its prepositional character. As such it is situated in a left-peripheral position of the whole infinitival projection, preceding the arguments (if present). This fits neatly with the fact that we have no attestations of spreading of verb doubling to other verbs in German-ALM. Brandner & Salzmann (2012) discussed only data from Zurich German and from the region around Lake of Constance. With the comprehensive data from SynAlm, the expectation is now that the form of the particle correlates with its position: those variants that use go/ga should have the particle in a low position within the VP, as a part of the verbal complex, whereas the others would posit it preferably at the left edge of the infinitival complement. This expectation is borne out, as Map 3 illustrates.11

Of special interest is the marked region north to Basel. In this area, High Alemannic is spoken. Recall that this is the sub-dialect that cross-cuts the political since hundreds of years. And as we can see, it indeed patterns more closely with the Swiss-Alm version.

A further indication that we are dealing with a parametric difference is the usage of gV as a preposition in some of the Alemannic sub-dialects. As mentioned above, there is consensus that gV evolved out of the common Germanic preposition gegen (>gen). The special property of gen is that it occurs only with place names. According to the DWB,12 it was common until the 16^th century and has since then been gradually replaced by the prepositions auf or nach:

However, in some sub-dialects of Alemannic, the preposition gen has survived and is used actively by the speakers. Map 4 shows the acceptance of gV in Alemannic preceding a place name.

Note first that the area in which gi is possible as a preposition cross-cuts the established distinction between Middle Alemannic and parts of Swabian while on the other hand in Switzerland, this usage is virtually non-existent. Such a pattern is typical for lexical variation and we will see more examples of it below. But here, it additionally correlates clearly with the (im-)possibility of reanalyzing the original preposition to a verbal head. Although I do not have area-wide data, informants from this region mentioned several times explicitly13 that they can use gi/ge only with intransitive verbs. Translated into structural terms, this means that these speakers only use gi/ge only with a nominalized form of the verb, without any functional structure. In the northern part of Swabian, gi is not used as a preposition and gi can occur at the left periphery of a more expanded infinitival complement to a higher extent. In the dialects in Switzerland and crucially also in the High Alemannic part in Germany, the prepositional use is not possible. The category label in these variants has eventually changed from P to V – with the relevant consequences for the syntax, specifically that it occurs left-adjacent to the lexical verb, forming a verbal complex.

In sum, the just-sketched case of syntactic variation shows all the traits that we would expect from parametric variation in the sense that there is a difference in the functional lexicon which has consequences for the outer syntax of the construction. Furthermore, the distribution in the relevant case concurs with the established sub-division of Alemannic.

As was discussed above, variation in the (non-functional) part of the lexicon is expected to show patterns that deviate in a more unsystematic way from the established boundaries. A lexical content item can easily be borrowed and integrated into a language without touching the systematic parts of the grammar. Therefore, whether a certain item is borrowed (and actively used) is up to the individual speaker in the end. And whether it spreads or not is an issue of socio-linguistic questions and thus again outside of the grammar. The lexical item that illustrates such a situation is the Standard German verb verwenden (utilize).

The task was to translate the sentence in (6):

Although verwenden is a very common verb in German, it is felt by many dialect speakers to belong more to the written style. And indeed, the informants replaced this verb to a large extent. The bulk of replacements consisted of either brauchen (utilize) or nehmen (take). The areal distribution is depicted in Map 5:

187 informants out of 529 (35%) used verwenden: the map reveals that the percentage is higher in German-ALM. However, this aspect tells us merely that this verb is transferred quite easily into the dialect – if it is offered in the stimulus sentence.

More interesting is the sharp division between brauchen and nehmen. Brauchen is used exclusively in Swiss-ALM. Even in the High Alemannic area in Germany, which patterned together with Swiss-ALM in the gV-construction, this lexical item is not found. Note furthermore that there is no single occurrence of brauchen in Vorarlberg, a region that also patterns usually more with Swiss-ALM. But note the asymmetry: verwenden occurs in Swiss-ALM as well as nehmen. The areal distribution of the latter – most of the occurrences are found near the border to Germany – suggests that this is due to language contact. Thus, Swiss-ALM speakers show interference effects for both verbs. On the other hand, German-ALM does not use brauchen. Thus, the conclusion is that brauchen in this context belongs exclusively to the lexicon of speakers of Swiss-ALM. Compared to the maps that were discussed in the previous section, the historically grown sub-divisions of Alemannic are obviously not the decisive factor.

This situation is expected if the considerations from above are on the right track. Lexical change or borrowing produces areal patterns that do not coincide with the established sub-divisions. And this is due to the fact that the lexicon is consciously accessible by the speakers and thus a matter of free variation.

In this sub-section, I will discuss an instance of the situation described under 2. in the introduction. This pattern is somehow in-between, as it does not follow the traditional sub-division, but nevertheless shows rather clear areal distributions in terms of clustering of certain variants. The phenomenon in question is the order in 3verb-clusters. The variation attested in 3verb-clusters (either in German or Dutch dialects) is one of those topics where keywords like “surface variation”, “optionality”, and “under-determination” show up regularly. The reason is simple: nearly all possible linear orderings of the three verbs are attested in the various West Germanic dialects, see Schmid/Vogel (2004) and Wurmbrand (2006, 2017) for a detailed and comprehensive overview about the data and theoretical approaches in recent years. Seiler (2004) detected an interesting areal pattern of variation in Swiss-ALM: he shows that the attested patterns can be modeled by assuming that head-finality increasingly vanishes from East to West. Such that the harmonic 123 orders14 occur only in the West. He shows that there is an inclusiveness relation and models it in an Optimality theory framework. Another prominent work on this issue in the context of the discussion here is Barbiers (2005). Confronted with a comparable situation in the Dutch dialects, he concludes that the actual choice of a specific order15 in a given area is not driven by the grammar, but must instead be delegated to outer syntactic factors. Finally, Sapp (2011) argues for the same point on the basis of diachronic data and shows that the factors that may have influence on a given choice of order are all outer-syntactic and thus that the variation cannot be captured as parametric variation.

With this background, let us have a look at rather recent data from SynAlm. The data I will discuss in this section are actually a “by-product” of a translation task that originally aimed at the form of the complementizer in temporal clauses. The sentence of this task is given in (7), (FB2_Q_12-3).16

Alemannic does not have a simple past; thus the modal verb inflected for past tense in Standard German was transformed by our informants into a finite auxiliary (have) and an infinitival form of the modal verb,17 giving rise to the patterns of the 3verb-clusters in (8) and compare with Map 6:

These were the only orders produced by our informants in this task. Since the original task did not aim at the order in verbal clusters, there were neither contrasting orders presented nor judgments or preferences asked. The informants produced these 3verb-clusters spontaneously and thus were completely unbiased.

First, Swiss-ALM shows a completely uniform behavior in that only the order aux-mod-lex (123) occurs. In the Northwest there is a clear preference for clusters of the 312 type, in which the lexical verb is in initial position but immediately followed by the auxiliary. The order aux-lex-mod (132), on the other hand, clusters in the Swabian area. Vorarlberg shows a preference for an initial position of the lexical verb as well. Recall that Vorarlberg belongs politically to Austria – a country in which Bavarian is the predominant variant. Bavarian does not show the pervasive predominance of verb raising as Alemannic or Dutch, although it is attested to a certain amount and in fact the 312 order seems to be the most common one, see Patocka (1997). The order lex-mod-aux (321) occurs exclusively in the northernmost area. This is an area that does not belong to the Alemannic dialect group.18

So far I have not discussed data from Alsace. The reason is that the number of informants is rather small and is thus not representative. However, the fact that the order in the verbal cluster is one of the few phenomena where the Alsatian speakers show completely uniform behavior is striking. If we take into consideration that Alsatian speakers are in general bilingual – with French as the other language – the strong preference for the 123 order does not come as a surprise. Recall that this is the only grammatical possibility in French. That means, we have two regions (Vorarlberg and Alsace) where obviously language contact plays the crucial role for the actual outcome of the order in a 3verb-cluster. This finding corroborates the analysis proposed in Barbiers (2005). Note that language contact seems to be at stake as well in the northern region where the 321-order “trickles down” into the Swabian area. 321 is the Standard German order and we can be rather sure that all dialect speakers are confronted with this version to the same extent. Nevertheless, the more to the south, the less the speakers produce it. Thus, language contact plays a role in the choice of the possible orders. But this is not due interference from Standard Germansince otherwise it would be hard to explain why this order is confined to a certain area. Recall that the interference effect concerning the lexical item verwenden did not show any regional clustering of this type.

I will refrain from discussing the syntactic analysis that produces the different variants but refer the reader instead to the above cited work. Important for the discussion here is that the choice of the order is prone to the linguistic environment and thus we can classify this type of variation as conventionalization.

A final remark on Swiss-ALM. That the Swiss-ALM speakers uniformly produced the 123 order seems to stand in a slight contrast to the findings in Seiler (2004: 380). He reports that 312 orders can actually be found in the East of Switzerland. There are two things to consider: (i) Seilers data basis is much more fine-grained than the one reported here: he makes a distinction between judgment data and preferred variants. The data here in contrast stem from one translation task and are produced “spontaneously”.19 (ii) the informants of SynAlm knew that the questionnaire had been sent to all Alemannic speaking areas. And of course, dialect speakers know about the prominent features of their own dialect and to a certain extent about the differences with neighboring dialects. Thus, it is entirely conceivable that speakers in this task deliberately chose the variant that they knew was essentially unique to their dialect. The important thing for the discussion here is that it is exactly this area of the grammar where we we find the leopard spots pattern.

In sum, this section has shown that the three differing patterns of areal distribution can be brought together with three distinct types of variation, discussed in section 1:

In the next section, I will show that all the three patterns established until now show up within one syntactic phenomenon, namely the distribution of the infinitival marker zu. The occurrence of the three patterns will then be taken as the starting point for an analysis of the infinitival marker that is able to account for this fact.

In this subsection, I will briefly give an overview about the common views on the nature and development of the IM before we enter the discussion about the variation in the contemporary dialects.

It is a widely accepted scenario that the IM in the West Germanic languages evolved out of the allative preposition zu/to/te and grammaticalized into an infinitival marker, see Haspelmath (1989), also Paul (1920, §345). The general idea is that the notion of goal which is entailed in this preposition can easily be transferred to the notion of future and/or irrealis. Semantic bleaching of the preposition then allowed it to occur in contexts without a future meaning such that it is now even compatible with factive verbs like e.g. regret – which refer to an event in the past.

Ijbema (2002), in her detailed discussion and analysis of the role of the IM te in Dutch, looks into the diachrony of Dutch te and finds an astonishingly similar situation to the one in contemporary Alemannic. Again, the critical cases are verbs of the forget/try-class where the IM is just like in contemporary Alemannic optional. Ijbema models this optionality by assuming that it is a side effect of the not yet completed grammaticalization of the preposition te into an IM. According to her analysis, te first realized the [irrealis] component within the IC. The bleaching mentioned above is accompanied by re-categorization to a particle and this made it possible for te to occur in T. This is the syntactic position where it is located in contemporary Dutch. Transferred to the other Germanic languages with a similar development, this scenario is much in line with the work on Tense in infinitivals. Stowell (1982) already suggested to posit a tense-node within infinitives, also Enç (1987), Landau (2004), but see Wurmbrand (2014) for a different view which will be taken up below.

Still considering the diachronic issue, there is abundant evidence that zu-marked infinitives increased during the history in all the West Germanic languages. Traditionally, it is assumed that the increase of zu-infinitives is due to a stabilization process in which the bare (or optionally marked) infinitives were replaced by zu-infinitives. But Los (1999) shows for English that the increase of the IM is rather due to the fact that the finite subjunctive complements of e.g. propositional attitude verbs were replaced by the marked infinitive and that the bare or optionally marked infinitives were affected only to a rather low degree. Smirnova (2016) presents a detailed diachronic investigation of manipulative (force, compel) and directive verbs (order, advice). First of all, zu-infinitives are found with these verbs already in Old High German, see also Demske (2001), whereas bare infinitives occur rarely with these verbs. Second, the spreading of the zu-infinitive is again due to the replacement of the former finite clauses in this environment. Thus, she confirms Los’ (1999) findings for German, namely that zu-infinitives replaced finite/subjunctive complements and not bare infinitives. The seemingly grammaticalization process consists thus merely of a higher percentage of zu-infinitives in these environments. The spreading of zu-infinitives to additional environments, due to the grammaticalization path postulated by Haspelmath (1989) and Ijbema (2002) and triggered by semantic bleaching of the preposition, can therefore not be confirmed. These findings cast doubt on the grammaticalization scenario sketched above, especially that the increase of zu-infinitives is connected to the semantic bleaching of the preposition, since it could replace the subjunctive embedded clause from the beginning on, even under verbs that should show this behavior only in later stages, if the semantic bleaching is a crucial factor in the development.

Considering in addition the contemporary variation with respect to the absence/presence of the IM with the forget/try-class in Alemannic – which is found in Middle Dutch and in Old and Middle High German as well – the conclusion is that what has been stable for hundreds of years is the variation concerning the presence/absence of the IM with the complements of forget/try-class of verbs. In other words: we seem to have detected an area in the grammar where underspecification is at stake in the sense that two different realizations are possible in nearly equal terms. And this variability23 seems to be systematic, given the persistence of the optionality of the IM. Therefore, one should seek a principled reason why this verb class is so special. As a final remark closing this section, recall that in standardized German only the zu-infinitive is licit, cf. (11). This requirement is probably more an artefact of the standardization process itself. During this process one of the two possible versions was chosen as the “correct” one for the written standard variant. Under this perspective, we are dealing with a case of conventionalization, this time driven by normative grammarians. In sum, the diachronic evidence as well as the areal patterns in the contemporary dialect hint at conventionalization, as defined under 2 in section 1 and illustrated in section 2.4.

Looking at the different types of verbs listed in (10)–(15) with respect to the presence of the IM, what immediately comes to mind is that it is connected to the simultaneity of the two events described in the matrix clause and the IC. The temporal extension of modal and perception verbs clearly coincides with the temporal extension of the embedded event. In contrast, e.g. with a propositional attitude verb like promise, the two events, the one of promising and the one described by the embedded verb, can be temporally divided. If zu is situated in T⁰, the non-simultaneity can be accounted for simply by the presence of a T-projection in the IC with its own temporal references whereas there is no such functional layer above the lexical verbal projection in the other cases. But as we have already seen, the situation is more complex and such a simple dichotomy is not sufficient: on the one hand we have the forget/try-verbs with their inherent variability, as just discussed; on the other hand, purpose infinitives and tough-constructions do not fit into the simultaneity picture either. Note that a purpose clause for example may be adjoined to a nominal expression and its future orientation is completely independent from the temporal expansion and the lexical meaning of the matrix verb. Tough-constructions like this book is easy to read have more of a generic interpretation since the IC denotes merely a property and thus the question of simultaneity does not arise in the same way as with modals and other IC-selecting verbs. Therefore, a new perspective on the role and the semantic contribution of the IM is called for. Such an approach will be sketched in thext section, followed by an application of it to the single cases of ICs whose areal patterns gave rise to the suspicion that the uniformity of the IM and its clear-cut distribution in Standard German rather camouflages its diversity.

In the following, I will use the universal spine approach, as proposed in Wiltschko (2014) and preceding work by her with Ritter, e.g. Ritter & Wiltschko (2009, 2014).

The important insight from this work is that Tense – as we know it from the Indo-European languages – is just one instance of the possibilities to anchor an event. Based on a detailed analysis of data from various languages, which are reported to have no tense distinctions in our understanding, they suggest that instead of temporal values (past, present, future), deictic local expressions (proximal vs. distal) or 1./2. person vs. 3. person pronouns24 may be suitable to anchor an event as well. They posit the universal category Anchoring Phrase and suggest that its featural specification consists only of the basic feature [coin] for coincidence. Coincidence distinguishes whether an event is coincidental with a contextually or linguistically given anchor or not. As such, it captures already the above discussed notion of simultaneity in a direct way and we will see immediately that the system gives us the needed flexibility to account for the situation described in the previous section. A general structure of anchoring is given in (16) whereby arg_sit is an abstract representation of the two contents to be related:

In a finite clause, the arg_sit in the specifier position hosts the utterance time (Utt-T) and the arg_sit in the complement position represents the event time (Ev-T). The [ucoin] feature will get a value that relates the (in this case) deictic value to the event time, [+coin] in case of present tense, [–coin] in case of future and past. The finite verb will then show the respective inflection (m-marking in Ritter & Wiltschko’s terms). In a language without tense-marking, the Anch⁰-head hosts a lexical item (locative, person) that will deliver an equivalent basis for the computation of the plus or minus value of [coin], e.g. the proximal preposition for [+coin].

Embedded finite and non-finite clauses have an AnchP as well. The difference is that the situational argument in the specifier is in this case anaphoric (pro) and it is bound by the event time of the matrix AnchP:

The value of the embedded Anch⁰ of an infinitive is according to Ritter & Wiltschko determined by the lexical properties of the matrix predicate. This type of assigning a value is called predicate valuation and it means that the matrix verb determines the form of the infinitive (specifically whether it shows up with an IM or not) via lexical selection. They then suggest that simultaneous verbs like try and begin value for [+coin] whereas propositional attitude verbs as well as manipulative for example for [–coin]. Thus, they adhere to a pure lexical semantic approach. As for the role of zu, they remain silent. The way it is implemented, the system can deal again only with the strict dichotomy, for which we have seen that it is not adequate. Thus, a slight modification is necessary.

But before going into the further discussion, a few remarks concerning the notion of coincidence. It goes back to Hale (1986). Hale argues that this binary distinction is found in essentially all areas of the grammar and that it is the one basic distinction around which grammars of natural languages are built. Besides the temporal system (including aspect) and the pronominal system,25 the difference between locative and directional prepositions can be subsumed under this notion as well. Interestingly, the core example in Hale (1986) for the illustration of coincidence and non-coincidence is the difference between the prepositions to and at. At is the proto-typical representative of coincidence, other examples are by, in, along, over, past, through, and with, see Rapoport (2014) for further discussion. To26 on the other hand, due to its inherent complex semantics is an example of non-coincidence. Which kind of local relationships does to entail? First, there is a fixed location, i.e. the goal like the store in run to the store. Remember from above that this is the component that is taken as being responsible for the future/irrealis interpretation of the IC – which then subsequently got bleached. Secondly, there is in addition a path (motion) involved, namely from the (not necessarily defined) position where the path starts until the goal is reached. Thus, the preposition zu entails a temporal expansion. It is this latter component which I will take to be the relevant one for its suitability as an IM: only if there is a notion of temporal expansion present, coincidence with an event can be computed in a meaningful27 way. These considerations make it superfluous to posit a semantic bleaching process for this preposition in order to account for the cases where no goal (i.e. future/irrealis) is involved. And as was discussed in section 3.1, there are serious problems with this assumption in any case. Furthermore, recall that with verbs that take typically an IC (manipulative verbs), the zu-infinitive was already present in the oldest texts. If I am right, there was no slow belaching process, rather, the preposition was used for lexicalizing the [coin] value from the beginning on. And this is not due to the notion of goal but rather that zu (like de in Romance) lexically involves a temporal expansion. It is this meaning component which makes it suitable for lexicalizing the [coin] feature.

Now recall that in the Ritter & Wiltschko system the value for the [coin] is not necessarily bound to the verbal system. There are languages that use regularly prepositions in order to anchor the event. I see no reason why languages should not mix the systems and use for the embedded infinitival [coin] feature a preposition as the overt realization (m-valuation) of the respective value. And as just discussed, the preposition zu is very well-suited for this task since it brings in a temporal expansion which is necessary for computing the [coin] value.

The suggestion is thus that in the case of a [–coin] value in a finite clause with the sub-specifications for past and future, the Germanic languages use the verbal inflection and the Anch-head is realized as a functionally extended verbal head (=traditional TP). In case of an infinitival complement, a preposition is the m-valuation of the [coin]. This means we do not have to adhere to predicate valuation, as Ritter & Wiltschko do. Note that if predicate valuation is assumed to be responsible for the valuation, there is no way to account for the distribution of the IM in a systematic way. But if the Anch-head shows m-marking on its own, similar to a finite clause, predicate marking in the sense that it directly determines the form of the complement via selection is superfluous. That the [coin] value of the IC matches with the lexical meaning of the matrix verb is then a matter of a “compatibility check” rather than syntactic selection. This will give us the necessary flexibility when it comes to the forget/try-class.

Based on this, I will suggest the following valuation rules:

Note that zero-marking for the [+coin] value does also hold for finite tense: present tense, i.e. [+coin], in the Indo-European languages is equally not marked morphologically. This parallelism makes the proposal attractive, since such a parallelism cannot even be formulated if one sticks to outer-linguistic notions like goal etc. for the reason why zu is used in ICs.

The proposal is furthermore much in line with recent considerations on the role of tense in infinitivals in Wurmbrand (2014). Wurmbrand argues extensively that there is no syntactic TP in ICs, in contrast to the more traditional treatments of ICs. According to her, three different types of infinitives have to be distinguished:

The main diagnostic is the (in-)compatibility with time frame adverbials like tomorrow, yesterday, which are computed based on the reference time of the matrix clause. The first class allows these adverbials (Today, John expects to leave tomorrow) whereas the other two classes show restrictions. Specifically, the third class does not allow such an adverbial at all (*John tried/began to leave tomorrow) whereas with the second class, an adverbial is possible if there is an additional aspectual marking (John claimed [to have left/to be leaving/*leave at three]. As I am mainly concerned with the third class, whose temporal properties will be discussed in section 3.3.1 in more detail, I will only briefly mention that the two other classes can be captured quite easily in the system here: for future infinitives, the value is clearly [–coin] and the occurrence of zu is expected. As for class II., Wurmbrand shows that the temporal expansion of the attitude holder’s NOW and that of the embedded event do not coincide (there is either shifted past or an imperfective in which the attitude holder’s NOW is included) and thus again zu-marking is expected.

Concerning the infinitival marker, Wurmbrand (2014: 414, fn 8) explicitly remains neutral when it comes to its position and function in ICs. However, she states that zu is certainly “not a tense element”. In addition, according to her, it seems pointless to give it a unified semantics in view of the fact that it is compatible with so many different meanings, cf. the different types of infinitival constructions illustrated for German above. But with the considerations about its meaning contribution in terms of coincidence and the ingredients just proposed, the undertaking is may be not that hopeless.

In sum, I assume instead of a TP an AnchP28 with its [ucoin] feature to be valued. Note that the assumption that the head of AnchP is category neutral in the sense that it might host prepositional or nominal categories as well, allows us to posit the preposition within the clausal projection without the necessity to postulate a process of re-categorization to a particle. In this sense, we can again depart from the traditional grammaticalization scenario. In the next section, I will now discuss how the three different classes of verbs for which we have seen the varying areal patterns can be analyzed in the system proposed here.

For the sake of completeness, let me just mention that the complements of modal and perception verbs now are expected to not show an IM. From a semantic point of view, it seems uncontroversial that they cannot have a different temporal expansion than the event described by the VP. And as mentioned above, throughout the diachronic development as well as in contemporary Alemannic, the ICs of these verbs are bare infinitives.

As a reviewer points out, the system proposed here, in which the IM is indeed meaningful, seems problematic for a verb like wollen. This verb construes like a modal verb with a [+coin] value and thus without an IM. The verb wünschen (wish) however has essentially the same semantics but construes as a future infinitive with an IM and thus has a [–coin] value.

First note that wollen is the only modal verb that can alternatively combine with a finite clause in Standard German:

Such an alternation is not possible with the other modal verbs. This shows already that the modal wollen has entered the lexicon of German with a different entry, namely as an alternative to wish/desire. Of course I admit that these verbs are quite close in terms of their lexical semantics. But this is not necessarily an argument against treating zu as a meaningful element the way I sketched it above. Close or nearly identical lexical content can often be expressed by using different syntactic constructions. Wollen is a modal verb that is very close to an interpretation as a future infinitive, especially since the ordering source does not come from the outside. Thus, it would even be expected to be able to occur with an IM – which it does not (yet?).

As the reviewer further notes, there is another problematic case where both types of ICs are possible: this is brauchen (need) in German. In its negated form, it is interpreted as a modal verb and consequently it can occur also without the IM:

Although I claimed above that modal verbs are temporally coincident with the embedded event, this is in fact not that obvious as it is for example with the perception verbs. The reason is that modalization brings in additional aspects than the pure temporal relation between two Anch-heads in terms of coincidence. Thus, I agree that a satisfying treatment of modal verbs requires a much more careful treatment in order to capture the different further semantic effects of modalization. It may very well be that the reason for their not having zu-complements must be found in fact somewhere else.29 However, again, this is then still not an argument against treating zu as a meaningful element. I will leave the discussion about these further aspects of modal verbs for future work. Instead I will turn now to the question how to analyze the simultaneous non-propositional class (try, forget) – exactly that class that shows the variable behavior systematically – and not dependent on different readings.

Consider first the following examples where in one case an achievement verb and an activity verb is combined with either a modal verb or try:

(20b) is out because the temporal adverb for hours is combined with an achievement verb, which does not have a temporal expansion. But when construing it with try, the adverb becomes possible. Adding a modal verb does not lead to this effect. This effect shows that, whichever further factors come into play with modal verbs, at least their temporal expansion is immediately dependent on the properties of the lexical verb, cf. the brief discussion in section 3.3. With activity verbs, no differences can be observed in either version. So the relevant contrast is between (20a) and (20c).

Grano (2011) suggests that try brings in an additional temporal component, which he calls the preparatory phase. And it is obviously this preparatory phase that can be modified by the adverb. The preparatory phase is seen as a (mental) state that turns into an inner stage as soon as the event described by the embedded verb has started. Importantly, when using try, the preparatory phase has already begun, i.e. one part of the complex event is realized already. This is the important difference to e.g. want where there is no such intimate connection. Want is a typical future infinitive in the sense of Wurmbrand (2014) which is compatible with two different time frame adverbials, i.e. yesterday, I still wanted to leave tomorrow. This is not possible with try, cf. *I tried yesterday to meet you tomorrow.31

A similar temporal structure can be posited for forget – although reversed in a sense since there is of course no preparatory phase. But still a mental state must be assumed that covers the inner stage of event the same way as with try. Thus, forget itself also has a temporal extension of its own – independent from that of the embedded lexical verb. When considering phase verbs, similar considerations apply and the same is true for dare. The important point is that in contrast to modal and perceptions verbs, there are two distinguishable phases with these verbs, but wshich are nevertheless temporally tied to each other: one where the event and try-state coincide and one where there is only a try-state in its form of the preparatory phase. The claim is now that exactly this special property is the basis for the optionality of the IM with these verbs: both specifications of the infinitival complement are compatible with the special temporal properties of these verbs – and thus it is expected that both versions occur to nearly the same extent.

Let us then come back different to the areal patterns. We have just seen that the overall temporal organization of the forget/try-class is inherently flexible. Thus, the point where variation from a semanto-syntactic point of view is to be expected could be identified. The question now is: how can we explain that in some areas the variability is directly reflected in the acceptance rates32 (recall that 39% in German-ALM accept no IM with forget), whereas in others, only the version with an IM is possible, cf. the Highest Alemannic region? Assume the following scenario: as we know from the diachronic data, the variation between IM and zero-marking with these verbs is a stable property of the ICs of these verbs. Thus, the language acquirer was and is confronted with variable input. But due to the special semantics of this class of verbs, either value for [ucoin] is compatible with the temporal organization and the optionality of the IM is tolerated by the grammar. Under this perspective, the question is rather why does the IM-marker occur so regularly in the Highest Alemannic regions, i.e. why do we not find the same amount of variation there? If we consider the maps for the forget-type, it is obvious that contact plays a role, i.e. the closer a Swiss German region is to Germany, the more the zu-less infinitives are accepted. Now note that the account proposed above does not per se exclude that there is a uniform marking with these verbs – even without a prescriptive grammar. In High Alemannic, the zero-form was obviously less common at a certain point in time. The variational input seized and the possibility of a bare infinitive finally died out.

In sum, the variational pattern that is found with the absence/presence of an IM in the forget/try-class of verbs is a typical instance of conventionalization in the sense that both values are compatible with the temporal structure of these verbs; in some areas, this situation is reflected directly by the near free variation between the different versions – in others, a uniform pattern is preferred – but there is no indication that either choice has further consequences in terms of interpretation.

Let us now look at the zu-less infinitives in purpose clauses. First recall the facts: the possibility of having a zero-marked infinitive covers by and large the same area. In German-ALM there is a high amount of variation, i.e. speakers can choose rather freely whether an IM occurs or not. In Swiss-ALM on the other hand, the area in which there is free variation is confined to the East. At first sight, this is a nice corroboration of the varying input scenario that we have seen above for the forget/try-verbs, as language contact seems to play a role. But we will see immediately that things are different here and that we are dealing with a “well-behaved” case of syntactic variation such that the possibility of variation is tied to the functional specification of the various complementizers used in a given variant.

A purpose infinitival clause is not selected by the matrix verb. Instead, it modifies either a nominal expression (I took x in order to achieve something (with x)) or it modifies the event expressed by the VP (I did x in order to achieve something (via x)). Thus, we might ask if the issue of temporal relation and thus the [±coin] is relevant at all in these cases. But clearly, purpose clauses occur with ICs and the reason is that the notion of purpose already entails that the purposed event is always a (relative) future irrealis – irrespective of the lexical meaning of the verb in the matrix clause. Now given this, the value in a purpose infinitival clause is always [–coin]. For this reason, they should always show up with an IM – contrary to the facts found in Alemannic. Recall that about 20% of the speakers accepted a zu-less infinitive in these constructions.

Now purpose infinitives are distinct from other ICs not only in that they are not selected but in addition, they are introduced by a complementizer; in Standard German, this is um:

In Alemannic purpose clauses may be introduced by um as well, but there two alternatives to this complementizer, namely zum or für (for English: in order). The Map 1033 shows the distribution of these three34 forms.

The region that uses für is co-extensive with the one where zu-less infinitives are disfavored, cf. Map 8. Thus, these two properties must be related and we should be able to find a syntactic explanation for the differences.

Looking at the numbers for this task, as summarized in Table 2, the correlation is even more obvious:

The table summarizes the varying patterns that were obtained from the translation task in FB3-6-3 and which is given in (22). The translations are categorized according to the element introducing the purpose infinitive (für, zum, um) and additionally according to whether or not the infinitival verb is preceded by an IM. About 22% with zum in the left periphery accept36 a bare infinitive. But less than one percent of those using für or um can have a bare infinitive. Thus, the choice of the initial element plays a crucial role in allowing a bare infinitive or not.

Note that the IM in these cases is zum. I will follow Postma (2014)37 who suggests that this form is the result of T-C movement of zu which adjoins to um which is in turn base generated as a complementizer in C⁰. This yields in the end the surface form zum. This analysis immediately gives us a clue to account for the attested variation: in those variants using zum, the [ucoin] feature is valued by zu in the Anch-head, lexicalizing the [–coin] value. The subsequent movement to C⁰ as one option yields then the surface form. If on the other hand a variant chooses für as the lexicalization of purpose38 in C⁰, the IM zu is indispensable to value the [–coin] value.39 Note finally that those speakers using um (the Standard German version) have the zu in the Anch-head as well. The same reasoning as for für applies: with um as the only functional item in the representation of the clause, the [–coin] value would not be lexicalized. As such, the seemingly optionality of the IM in purpose clauses is of an entirely different nature than that with the forget/try-verbs. It is a clear case of parametric variation.

In sum, the discussion of these two infinitival constructions has shown that the possibility of accepting bare infinitives is the outcome of several factors, namely:

These two cases show how important it is to consider the areal variation. The pure numbers as in table 1 would not have led to the findings just discussed. In the same vein, only considering the distribution on the maps would have lead one to categorize this variation as a further instance of conventionalization. However, if the two types of information about variational dialectal data are combined, the maps and the numerical data, then the more comprehensive data can lead us to more adequate analyses.

Let us then finally turn to the tough-construction. The example given to our informants was:

Recall that in this case, the areal pattern reminded on the pattern for lexical variation, illustrated in section 2.3: there is a sharp division along the political border between Germany and Switzerland. Whereas 26% accepted a version without the IM in German-ALM, this version was categorically rejected in Switzerland. I claimed that such an artificial division seems to be typical when it comes to lexical variation. So what is the difference between tough-constructions and the other infinitival construction? For reasons of space, I will not go into questions that touch on the suggested syntactic derivations of tough-constructions, see for example Hicks (2009). I will merely consider those aspects that are relevant for the temporal organization. First, as with the purpose clauses, there is again no direct interaction with the lexical semantics of the matrix verb and the infinitive. This of course has to do with the fact that the only type of matrix verb that occurs in tough-constructions is the copula. A copula merely asserts that a given property holds at a certain time. Under this perspective, the IC in this case denotes a property rather than an event. Thus the question of coincidence is in a way obsolete since there is trivially always coincidence. In the system proposed here, this would mean that we do not expect the IM zu at all. Still, in Standard German, it shows the IM zu, indistinguishable from the IM occurring in the other types of ICs.

Again, a more fine-grained analysis of the data is necessary. Until now, I neglected the issue of the form of the IM. Its typical form is zu (which is realized as z’ in Alemannic). Another form that we have encountered is zum, e.g. in the C-position in the purpose clauses. In Swiss-ALM, zum in purpose clauses occurs only in the C-position, as is evident from table 2. But note that in German-ALM, 52 informants used this form40 even in the low position inside the VP, i.e. it seems to be a phonological variant of zu. This impression is bolstered by the observation that the form zum may occur in German-ALM even in ICs under the forget/try-class (acceptance rate 32%) whereas in Swiss-ALM only 6%41 of the speakers accepted this form. Furthermore, zum in tough-constructions is highly preferred (80%) in Swiss-ALM whereas in German-ALM, either z’ or zum or – as seen – even the zero-variant may occur. From this picture we can conclude that zum in German-ALM can be conceived of as a lexical variant of zu whereas in Swiss-ALM, it is restricted to either the initial element in purpose infinitives – where it does not have a pure temporal interpretation – or to tough-constructions. Under this perspective, we are indeed dealing with a case of variation on the lexical level and the unexpected pattern in map (12) could thus be subsumed under the label free variation. But only in the sense that in German-ALM, the surface form of the IM is manifold, including a zero-variant. This would mean that the nature of the zero-marking in tough-constructions is again of a different nature than the variation with the forget/try class respectively with purpose infinitives. In which way the use of the zero-variant in German-ALM indeed correlates with a different syntax must be left open here. Until now only the data reported here are available. Still, the weird pattern with the tough-constructions should lead us to doubt that the seemingly uniformity of tough-constructions with the other types of infinitives in Standard German (and in English) are an indication of a similar syntax.

The data from Swiss-ALM show clearly that the IC in tough-constructions is of a different nature. What immediately comes to mind regarding the form is that the IC in tough-constructions is nominal in nature since the preposition bears an inflection for dative (-m); equivalent with occurrences as in Ich geh zum Arzt (I go to-dative doctor). In German-ALM, this specification seems to be lost (or can be cancelled) as zum can occur in verbal environments. Similar considerations would apply to other occurrences of the preposition zu with a non-finite verb form as in the equivalents to English something to read or Dutch zit te lesen (sit down to read), the latter with a progressive interpretation. It seems again that the non-finite verb in these cases shows rather the traits of a nominal structure rather than a clausal projection. If this is true, then there is no Anchoring Phrase at all. On the other hand, progressive and the modal ability-reading (as in something to read) clearly denote a temporal expansion. Future work in dialectal research should have a close look at the maybe very tiny morpho-phonological differences in the realization of the preposition in these cases. It is highly plausible that similar differences as those found in Alemannic will be detected in the dialects/spoken variants of English and Dutch as well.