Intro

"Why don't you open the door?". To express the fact that that the answer involves an ongoing activity, languages such as English or Dutch require an auxiliary cum (gerundive) participle construction whereas languages such as Czech or French don't:

What does that tell us about language, and the particular grammar of these languages? Why can't English use the synthetic form, or Czech the gerund construction?

In the habitual reading, a context such as "what do you do when you're bored?", both English-like and Czech-like languages use a synthetic verb:

The features expressed in a habitual construction and a progressive construction are different, as witnessed e.g. by the fact that they are lexicalised by different function words/morphemes in English. The fact that they come out identically in Czech or French is thus a classic case of syncretism:¹

The original question now turns into: what does it tell us about language that these two readings can be syncretic (and often are), and how does a particular grammar encode the fact that these readings are syncretic or not? These issues become both more salient and more revealing with the discovery of a new tense with varying syncretisms with the habitual and the progressive.

The auxiliary + gerundive participle shape of the construction leads us to expect a single ongoing event, but the meaning is that of a habitual: within the 'lately' timespan, it is usually/habitually/normally the case that Salvador doesn't eat well. The same 'habitual in progressive clothes' is found in the past tense:

And in the perfect:

An important difference with respect to the habitual is that the timespan must be relatively restricted:

If the temporal modifier is omitted, it is understood as a timeframe not too distant from the reference point:²

No such restriction holds of the habitual, which is readily interpreted as life-long or extended properties:

I will thus call this new tense the "medial habitual", to highlight its medium distance in time to the reference point.

Given its syncretism with the progressive, English can be characterised as an AAB language: the progressive is syncretic with the medial, both of which differ from the habitual. In informal registers of Dutch, speakers follow the same pattern as English:

More formal registers of Dutch, and all registers for some Flemish speakers, require the morphosyntax of a habitual in order to express the medial semantics:

This variety of Dutch thus has an ABB pattern: the progressive is alone in its morphosyntax, with the medial and the habitual syncretic. On a structural interpretation of syncretism, the fact that the medial can pattern either with the progressive or with the habitual indicates that it has a syntactic structure intermediate between them:

This predicts that the medial tense in Czech/French-like languages will be syncretic with both the progressive and the habitual: in those languages the medial is sandwiched between two syncretic tenses, and because of the *ABA pattern must be syncretic with them, which is indeed the case:

Composing the habitual into a progressive

Given a structural approach to syncretism (e.g. Starke 2002, 2009, Caha 2009, 2020, Vanden Wyngaerd 2018, Wiland 2019, De Clercq 2020, Taraldsen 2021, etc.), the simple syncretism between a habitual and a progressive entails that one of them is built from the other:

Under (18a), a habitual is simpler and a progressive is more complex:

The reverse holds under (18b):

A surprising consequence is that the meaning of the progressive must be compositionally built from the meaning of the habitual by adding (at least) one feature, or vice versa the meaning of the habitual must be compositionally built from the meaning of the progressive by adding (at least) one feature. How could that be? Barring features denoting arbitrary functions, it is far from clear how these compositions could work. Adding the medial however suggests a new way of looking at the meaning of the progressive. With the medial included, the underlying structure must be one of:

Either of these would work to derive the distribution of morphemes and auxiliaries. I will tentatively proceed with (21a), based on two informal hints: first, the mechanics come out simpler with (21a); second, habituals are typically morphologically simpler than progressives, where they differ – that is expressible under either structure above but comes out more naturally under (21a). The three tenses are now underlyingly:

The progressive meaning is thus created by adding two features to the meaning of the habitual – how does that work? Let us briefly explore an intriguing possibility, based on medial habituals. In (22b), the meaning of medials is built from the meaning of habituals by adding y. We saw above that medials have essentially the same meaning as habituals, with an additional restriction to a limited time span. It is thus natural to think of y as introducing a time span restriction.

If the build-up from x to y consists in adding a time restriction, a natural semantic continuation would be to restrict the time span even further: from unrestricted to medial to proximal. What would be the interpretation of a very restricted time span, in fact so restricted as to fit only part of one event? The event would 'overflow', yielding a not-yet-finished, i.e. ongoing, reading.

To put it differently, it may be that the 'essence' of a progressive is not in the progression, or ongoing-ness of the event, but rather in trying to fit an event into a time span too small for it, triggering an in-progress reading. Under this view, the higher two features might be more appropriately named along the lines of:

Regardless of the exact labels and interpretations of the three features involved, we now have a first answer to our original question. The differing patterns of habituals and progressives in English-like versus Czech-like languages reveal an underlying property of language in general: one of these readings is built from the other, syntactically containing it and semantically composed from it, as in (21), with a possible implementation as in (23).

In such cases, the higher features cannot lexicalise as they do not form a constituent. Repair movement kicks in, creating the following configuration (in several steps that I omit here for brevity, see Starke 2021 for a detailed walk-through of such derivations):

As a side-effect of this repair movement, the high features now linearly follow the evacuated constituent and hence end up being suffixes (or post-verbal particles). This is how suffixed roots are born.³

This mechanism however cannot give rise to prefixes, pre-root auxiliaries or pre-root particles. How do these arise? Since lexicalisation only ever applies to constituents, they must form a constituent independently of the root. Since they precede the root, the only way for their features to form a constituent is to be a (base-generated) complex left branch:

Under which conditions would such a complex left branch (CLB) be created? Creating a complex left branch requires starting a new sub-derivation, which we assume to be costlier than simply continuing the current derivation. This is thus a truly last-resort repair mechanism, when repair movements fail to create a successful lexicalisation. In such cases, syntax gives up on trying to lexicalise the feature itself, and instead builds a complex left branch containing that feature and uses it to label the main branch.

This then gives us the general shape of the answer to the language-specific question – why can't English use the synthetic form to express a progressive, why can't Czech use the auxiliary construction? Czech-like languages have lexical items capable of lexicalising the (relevant) features of a progressive with a bare or suffixed verb, and hence will not resort to the costly strategy of CLB formation. English-like languages don't have such lexical items, and hence the last resort strategy of creating a complex left branch (the auxiliary) kicks in.⁴

As standardly observed, the English word order suggests that tense is represented higher than the progressive aspect in the underlying structure:

The functional sequence around the progressive is thus at least:

By the same logic, passives show that the lexeme ed is able to lexicalise low features, below the progressive:

Let's call that low feature x. The overall functional sequence becomes:

Given this fseq, how are English habituals such as (33) lexicalised?

The relevant structure is now:⁷

We know that the morpheme ed will end up being a suffix, so the logic of suffixal derivations outlined above tells us that ed lexicalises the higher part of the structure and triggers repair-movement of the lower part (lexicalised by the root chop). This in turn means that the lexical entry for ed matches at least the highest feature of (34), T.

Recall that we also know that the lexeme ed contains the low feature x (based on (30)). We now have enough information to deduce the content of the lexical entry of ed: the only way for it to contain x, contain T and still be able to exactly match some part of (34) is:

This would yield a derivation of the type:

in which ed matches the right branch, and chop the left branch.

What happens when syntax builds a (past) progressive? The syntactic structure to be lexicalised is:

The lexeme ed is now unable to lexicalise up to T: it doesn't contain any constituent [ prox [ timespan [ hab [ x ]]]], as it doesn't contain the prox and timespan features. This leaves T stranded and unlexicalisable so far, with the progressive features blocking the path of ed, so to speak.

Could other lexical items realise T? The only other English lexical items containing past features are auxiliaries (was, had, etc.), here represented by was. Again, their precise internal structure is beyond the scope of this contribution, so let us simply make it parallel to ed:

Can this entry save the stranded T in (37)? Notice one important technicality: during the derivation of chopp-ed, the root evacuates, turning the higher part of the structure in a constituent independent of the root, and hence lexicalisable on its own. This has a surprising technical consequence: the lower-most layer of ed will only have one single branch (since the other branch evacuated), as in (35). In base-generated complex left branches, such as auxiliaries, the situation is the reverse: being base-generated, nothing has evacuated out of them, and hence their lower-most layer has the usual binary structure, as in (38).

It thus follows that the lexical entry for was cannot lexicalise any evacuation structure: its binary foot prevents it from matching the bottom of a derivation with evacuation. In English, a simple root+suffix derivation for the progressive, as schematised in (37), is thus truly doomed. No amount of rescue movement (or backtracking) will allow it to lexicalise.

English is therefore forced into the costly last-resort strategy of creating a new derivation, ie creating a new functional sequence in a new derivational 'workspace' – and hence a new verbal element – with the goal of providing a T feature. This is what we call auxiliaries. The ultimate culprit here was the shape of the lexical entry for ed: it lexicalises a past habitual, bypassing the intermediate progressive layers. When syntax builds a progressive, ed is blocked and nobody else can then lexicalise T short of creating an auxiliary structure.

Let us go over this last-resort auxiliary-based derivation in some more details. We first need to ask who lexicalises the progressive and medial layers, prox and timespan. The obvious candidate is ing. The lexeme ing has the further property of never co-occuring with the suffix ed, which suggests that they compete for the same starting point, the same foot (this not a necessary assumption, but the simplest for our purposes). This gives us the following entry for ing:

The derivation up to prox will be:

With ing lexicalising the right branch. The next step is to merge T. The structure cannot be lexicalised as-is, so repair movements are attempted, with successive cyclic movement first:

No English lexeme can however lexicalise the right branch, T is stranded, and that derivational path crashes. Accordingly, repair via complement-movement is attempted instead:

Again, no English lexeme starts at T and hence the right branch cannot be lexicalised. As a last resort, the derivation gives up on lexicalising T directly and instead spawns a new subderivation which attempts to lexicalise its own T and use it to label the mainline fseq. English does have a lexical item capable of lexicalising an entire T-containing subderivation, the auxiliary (38), creating a successful derivation:⁶

What is the situation in Czech-like languages? Recall that the habitual is syncretic with the progressive (and the medial habitual):

The progressive/habitual syncretism is below the features of the third person singular, here lexicalised by a thematic vowel (TV), and is hence plausibly done by the root itself.⁸ The lexical entry of such a Czech root would thus be along the lines of:

The suffix á then picks up above prox, lexicalising T and ϕ features (both T and ϕ are shorthands for a hierarchy of simpler features expressing the difference between present and past, or the various persons and numbers):

A habitual in a Czech-like language would come out as:

Progressives and medials come out essentially the same, here illustrated for the progressive reading:

To complete the picture, some languages do show suffixes or post-verbal particles in progressives. Here is Mandarin for instance:⁹

This would follow if the lexical entry of Mandarin ne was along the lines of:

with a derivation amounting to:

We were thus able to pinpoint the difference between Czech-like languages and English-like languages to a precise lexical contrast – English-like past tense morphemes gap the progressive whereas Czech-like don't, they start above the progressive:

Appendix: Prepositional progressives

Beyond the gerund-based construction and the simplex-verb construction, languages also express the progressive tense with prepositional constructions, sometimes involving a lexical noun. Take this French pair:

This ±prepositional difference comes with another difference: the synthetic and analytic strategy were in complementary distribution in English-like and Czech-like languages, whereas the two construction are (almost) freely interchangeable in French. Why is that?

The fact that a preposition and a lexical noun are involved, suggest that the syntactic structure being lexicalised in (53b) is different than the simple verbal functional sequence discussed above, and involved in (53a). The two structures are thus not alternative lexicalisations of a common substrate, and hence do not compete.

In fact, they seem to illustrate the difference between a functional projection and a full lexical concept-bearing lexeme. In (53a) the concept progressive is expressed via a feature merged into the English and Czech functional sequences, proximal if our semantic reasoning above was correct. In (53b) on the other hand, the notion of progressive seems to be contributed by the idiom (en) train, plausibly a lexical concept rather than a functional projection. If so, the notion of progressive is not grammatically encoded at all in (53b), there is no prox feature. The notion of progressive in (53b) rather comes via the conceptual content of one of its content-words. This is very much the same distinction as the notion of plural being grammatically represented by a syntactic feature and lexicalised by -s in bean-s versus the notion of plural conceptually part of the lexical word flock as in e.g. There was a flock in front of me with no grammaticalised representation of plurality, the notion being brought in via the conceptual meaning of the word flock.

Dutch progressives are a challenging middle ground: just like French, they are prepositional, and seem to involve a different syntactic structure than the verbal functional sequence of progressives discussed above:¹⁰

As a closing thought for this appendix, let me offer an intriguing possibility: Dutch simply has no way to lexicalize prox in a verbal functional sequence. Building a syntactic structure with prox would crash, and hence such structures are ineffable in Dutch. To work around this, Dutch needs to resort to an altogether different type of structure, the idiom aan het V-inf zijn lexically linked to the progressive concept. This creates the illusion of a grammatical complementary distribution: only the prepositional (idiomatic) construction is possible for the progressive, while the habitual is expressed by the default simple verbal functional sequence.