Tromsø Group Meeting: October 27 (Gawron)

Gawron's Generalized Paths

The Tromsø group discussed this paper over two sessions. We thought that the big idea in this paper was very clear and very compelling. We were convinced that the notion of path, in the traditional sense of mapping some varying property onto a time line thus expressing change, should be generalized to a mapping of that property onto a spatial axis, thus divorcing change from time. As a result of this idea, the group got into various discussions about the empirical data for different verbs and different languages seeking to explore the universality and limits of the phenomenon. The other aspect of our discussions centred on the mathematical implementation of the abstraction that would be required to make these facts fall out of a generalized semantics for the verbs in question. We found the paper very compressed and concise and spent a long time trying to unpack the slice function in particular. By the end of the second session, our empirical discussions and our technical concerns began to dovetail so a neat division was not always possible.

Event and Extent Readings by Verb and Language

We discussed mainly interesting observations from the different languages represented. Many of these observations were concerned with ways in which one can isolate one of the readings. For example, in Persian, the following pattern obtains:

(1) ...WIDE BECOME ambiguous
...WIDER BECOME stative only

In German, it seems that there are (many?) verbs that have either one or the other reading, but not both. for example, 'sich ausbreiten' (related to English 'broaden', and with the reflexive 'sich') only gives rise to eventive readings, and 'reichen von' (related to English 'reach') only gives rise to extent readings. The idea was floating in the air that perhaps reflexivity has something to do with this, but we didn't make progress on this idea. I guess partly because it wouldn't hold for other languages, such as Spanish (and I think also Russian), where verbs that have a reflexive particle can give rise to both readings. An example from Spanish:

(2) La niebla se extendia desde/de Londres a Paris
the fog reflx. extend.past.imp from L to P

It seems that both in Spanish and Russian the aspect of the verb can, perhaps expectedly, affect the availability of readings. In Spanish, changing the aspect of the verb in (2) to perfective ('extendio') results in only the eventive reading being there.

In the case of mapping a dimension onto a spatial axis, we discussed the observations that the dimension chosen as the spatial axis is subject to contextual manipulation, but that in general, one needs to choose a spatial axis whose measurement is a monotonic measure with respect to that object. Thus, the cable can widen along its length, but cannot lengthen along the dimension of width. This asymmetry seemed to be related to the Schwarzschild generalization about measures and nouns within the NP. So, one can say `one metre of cable' and not `one metre cable'; whereas one says `two inch cable' and not `two inches of cable' for referring to width.

We had some discussion of the different readings possible with `extent' verbs and the degree achievement verbs. Consider the sentences below.
(3) The fog extended from London to Paris.
This has the two readings laid out by Gawron as follows:
Reading 1: the fog moves and space covered by the spreading motion increases from including just London to also including Paris (event reading)
Reading 2: The fog statically covers the region between London and Paris (extent reading)
(4) The crack widened from the north gate to the south gate.
Reading 1: The crack gets wider as a whole, and the part of the crack that was getting wider is delimited by the two positions (event reading)
Reading 2: The crack is not moving, but the dimension of wideness is greater at the south gate than the north gate. (extent reading).
Gillian also detected a third reading for verbs like widen, if the subject was chosen correctly.
(5) The fog widened from London to Paris
Reading 3?: The fog moves from London to Paris, getting wider all the time (event reading, but tracking width as well)
Peter did not get this reading at all, and thought it was a very forced kind of coercion, but Gillian said she got it very easily. We wondered whether this third reading was possible for other speakers and/or other languages.
Whether the double reading exists or not, there are clearly two different scales that can potentially be mapped onto the time line: in the case of the extent verbs it is the material constituency of the subject; in the case of the degree achievements it is the increase of the property denoted by the adjective. If a degree achievement can be coerced into a manner of motion verb (possible for Gillian, but not Peter), then both scales seem to be mapped on to the time line. Gawron does not conceive of the extent readings as being mappings from stuff to time, but builds it in as a part of the notion of `spreading motion' as opposed to the motion of a rigid object. We wondered whether the material constituency scale could be unified with/analogous to Krifka's creation/consumption objects in event readings.

We noted that it is nice that in English the same temporal prepositions that are used to test for telicity (in the 'in'/'for' test) can be used to test for change in the space domain, as in the examples in (10) in the text. We noted that this is not so in all languages (e.g., in Spanish, 'for' is translated as 'durante' (similar to 'during') and cannot be used in the spatial case).
We did not this that this undermined the general point about `aspectual' properties being preserved in the generalized sense, just that English made the analogy even more striking.

Demon Mathematics

We found it hard to visualise four dimensions and so had trouble unpacking the effects of the slice function. One question that arose was whether the `demon' function in principle knew everything about all possible property and dimension scales, or whether it only understood time and three spatial dimensions for location. If the former, then it would be tracking in principle an almost infinite number of scales, only one of which would ever be picked out to intersect with the object's spatiotemporal trace function. Klaus wasn't convinced that the Tplane was intended to cover property scales like the ones required for degree achievement verbs since Gawron only talks about chosing S to be temporal or a spatial axis. Luisa and Gillian had read it as in principle generalizable to other scales. In any case, Klaus may have been right, since empirically there seems to be an asymmetry among scales: among verbs it seems to be possible to track any scale with respect to either space or time, but we couldn't find cases of where something was tracked with respect to a non space-time scale. If periphrasis is employed as in (6) , this kind of tracking seems to be possible, but we couldn't find a grammatical use of a verb plus the relevant path phrase to get the same effect:
(6) Stars get redder as they get cooler..
(7) *Stars cool from x wavelengths ps to y wavelengths ps.
(8) *Stars redden from x degrees to y degrees.
(For these judgements, we had to abstract away from the event reading, where an individual star or stars is changing in its property over time. These seem possible, but parasitic on the time line. The readings we were after were those in which lots of stars are being looked at and as we move on the colour scale we find stars that are different in temperature, and vice versa. These static readings for (7) and (8) seemed impossible to us.)
We surmise that time and space are privileged as the basic axes along which other things are tracked.

Another question that emerged was the use of the START and END functions in the description of the increase function found in verbs such as `widen'. (We assume incidentally that this comes from Hay, Kennedy and Levin 99 which are are going to read next in any case). The issue that arose here was whether it was enough to simply state that the degree of wideness was greater at the end of the event than at the start. Gillian's intuition was that the condition was stronger and that wideness had to increase monotonically with whatever the axis was (either space or time). Peter disagreed with these judgements and the data proved harder to disentangle than we imagined. First of all, one needs to recognise and abstract away from the granularity issue--- something can be perceived as generally getting wider even though a low level inspection of the measurement would surely show internal fluctuations. Secondly, the intuition about whether gradual monotonic increase was required seemed to vary with the particular kind of PP chosen. (spatial readings intended in 9-11 here, but the same point would be made with parallel event readings for widen)
(9) The cable widened from the living room to the basement.
(10) The cable widened in the basement.
(11) The cable widened along the skirting board.
Peter got a `just has to be wider in the basement' reading for (9), while Gillian got a gradual increase with no major reversals reading for (9). For (10), Gillian got either (i) suddenly gets to be wider in the basement or (ii) `starts to gradually increase in width as you hit the basement' reading. Peter also saw the ambiguity in (10) but thought that the (ii) reading was irrelevant. However, for both of our English native speakers (P and G), the gradual monotonic reading of `widen' seemed to be forced in (11). It's hard to know exactly how to interpret this data. Peter was in favour of retaining START and END, but allowing pragmatics and/or the semantics of the PPs to do the rest (possibly by enforcing `increase' at all subintervals to get gradual readings if one needed them). Klaus thought that it was better to build in monotonicity with respect to the scale in the increase function from the start (since it was easier to `coarsen' up structure than interpolate it). Gillian agreed with Klaus, but we had no way of proving things one way or the other at our meeting. We hope to get further on this question when we read the Hay, Kennedy and Levin.

Tromsø Group Summary: October 6

The Tromsoe group had its second meeting on Thursday (6th October) to discuss Marcus Kracht's paper on the semantics of locatives. Once again, discussion was lively and, we think, focused on slightly different things than the Utrecht group focused on. Here is a brief summary (from Gillian and Luisa's perspective).

THE LINGUISTIC REALITY OF LOCATIONS

Most of us were already convinced about `the linguistic reality of locations', but there were some elements of the discussion here that really gave us some extra food for thought. Treating a simple NP/DP as denoting an object or a place seemed to have linguistic and morphological ramifications beyond simple pragmatics. Gillian brought up the general point that a DP such as `book' can be used in a variety of different ways, e.g. to denote the physical object, the type of object, the content of the text etc. without much overt linguistic signalling. Changing mode in this sense does not even preclude using the same pronominal form:
(1) As for that book that John just put on the table, Mary has already read it. (not necessarily the same token).
One might think that the difference between referring to a location qua object, or referring to it qua location would be much the same phenomenon as the well known one above. It was interesting to discover that in many languages that wasn't so, but that the distinction had grammatical consequences. Even in English one might find hints of this. So, for example, in a recent paper on names, Ora Matushansky points out that while acronyms generally don't use the definite article (AIDS, UNICEF), abbreviations seem to need it (the CIA, the FBI, the RSPCA). One consistent exception to the need for `the' was the class of geographical and institutional locations (MIT, UCSC).

Another thing this reminded us of was a small class of location words in English that are sometimes used in complex prepositional forms (Peter talked about these in his seminar on prepositions a couple of weeks back). I am thinking here of things like.
(2) in front of the car; in back of the house (American); on top of the table.
Here, unusually, a definite article is not required either. This looks like a special group of localizers (the kind of thing that a place name, or a work like `Beijing', or `train-station' in Chinese might already incorporate syntactically).
Gillian thinks that this kind of `localizer' should be kept distinct from the `place' prepositions such as in, on, at, etc. When all three pieces (path, place and localizer) exist they seem to come in this order:
(3) to in front of the house
Path [Place [localizer (linker) GROUND OBJECT]]]


TECHNICAL ASPECTS: BUILDING IN TIME;REGIONS VS NEIGHBOURHOODS

We weren't convinced that one needs to make locative expressions sensitive to time, or at least that we shouldn't build it in to the denotation of the locative itself. This seems to us to beg important questions with respect to compositionality. The facts seem to be that extended locations are parasitic on verbal or tense material for their path-like or temporally ordered interpretation. We would prefer to think of paths as embodying some kind of concatenation/subpart relation (as in Zwarts' system) which is then homomorphically mapped onto some temporal scale (possibly via a homomorphism with the event subpart relations). We weren't convinced by the argument based on examples such as "Peter drove in front of John's car" . We think that this has the syntax/semantics of a static event description where the frame of reference is relativized to the (more or less) constant speeds of the two cars. Doing it any differently would be tantamount to trying to build in the rotation of the earth to statements about `the book being on the table'. (Actually, its a cool example, and its interesting that we can do this as speakers. It says something about the importance of contextual frames of reference.)

There was a brief discussion about the relationship between locations and participants as opposed to that between locations and events. Gillian actually agreed with the intuition that there was a deep difference here (which formalisms don't often reflect). The way she put it was that individuals are independently individuated entities which can then be `located' as one of the properties that can be true of them. Events on the other hand, do not have independent identity criteria---- they are not independently identifiable previous to being located. In fact, their locations are in some sense criterial of their identity.

Concerning the argument that we really need to be looking at neighbourhoods as opposed to simple regions, we basically had a discussion that replicated the Utrecht group's summary of this point. The notion of `touching' seemed crucial, but we thought we could rule out the hovering bird without making reference to neighbourhoods. The trick of course is how to exclude the bird from being `auf' when the top book on a very high stack still counts as being `on' the table (at least in English, according to Gillian's intuitions). Is this also true for German `auf' ? In any case, some fancy pragmatics is clearly needed here and its not clear to us that neighbourhoods help.

SELECTION AND THE EMPTINESS PRINCIPLE

Finnish is of course super cool, and the empirical generalization concerning the internal structure of locative cases was compelling. We had a discussion about the compositionality of the semantics for a language like English, which sometimes seems to show both M and L overtly, but we could not decide whether the semantics we got was the same as reported for the compositional M [L cases in the Finnish system. For example, in English (4) below, the path of motion never needs to change from being `inside'/`outside' the house as long as it starts at a location that could be so described.
(4) They crawled slowly from inside/outside the cave.
This shows that `from' in English does not necessarily create a path that is `diphasic', even though it is coinitial. We assume that the Finnish suffixes do need to be diphasic, but we had no Finnish speaker on hand to check this for something like talo-s-ta. Zhenya, who knows some Finnish, reported that if you really wanted to say someone came `from inside the house' you do use a full prepositional form (but once again, we didn't know what the judgements were).
(5) sisa\lta\ talosta
inside-ABL house-ELA
(from-on) (from-in)
We pondered the conundrum of the Ablative case on the `inside' word for a while just because we couldn't help ourselves, but then we gave up, noting merely that it would be interesting to know whether there were any semantic differences between full prepositions in Finnish and the suffixes (apart from lexical encyclopedic content of course).

We also had a long discussion about the notion of selection and the emptiness principle, which we didn't find satisfying at all. From an empirical point of view, Gillian pointed out the fact that it actually makes some semantic sense for a verb like Finnish "forget"/"leave" (I use the English words here) to always combine with cofinal and for Finnish "find" to combine with coinitial. `Forget' is a transition verb, but the location of the forgotten object is as described after that transition. With `find' it is the opposite: the stated location of the object is where it would have been before the transition. This seems suspiciously logical to us.

Of course, it is always technically possible to build in the semantics of the case form into the verb when the two always go together, leaving one with the illusion that the case form itself is vacuous in these cases. But this still leaves us with dumb `syntactic selection'. We think the difficulties of making syntactic selection work (especially for specific lexical exponents of a category head) are massively underestimated and very seldom worked out by the syntacticians in a satisfactory way. Syntactic selection is the problem, not the solution.

There are three options as far as we see it: (i) syntactic selection for something on the level of a category feature that can be checked in syntax (ii) semantic selection via unification of two regular and understood semantic bits in a compositional fashion; (iii) phrasal `idiom' formation. It is clear that (i) won't work here. Gillian would prefer to at least try for (ii). Even if (iii) were correct, there is still a lot to be explained since the idiom in this case would be a non-constituent, since it would exclude the nominal at the base of the M[L []] structure.


Furthermore, the memorized collocation account seems to suggest that the choice of mode, or mode+localizer is totally arbitrary -- it can change from verb to verb, and it can change from language to language for the same verb. For example, we don't think it is an accident of cross linguistic selection that in English and Indo-European static mode is chosen for a certain class of punctual verbs, whereas in Uralic languages and Finnish the selection is for `cofinal'. Treating it as an accident seems to be missing generalizations, in particular about the verbal functional and morphological structure of those languages. The selection account also misses generalizations concerning cross linguistic similarities. For example, it seems to miss the fact that many languages, not only German, choose things similar to 'vor' to combine with "afraid" , which seem to have some semantic rationale behind them.

The discussion was based on a handout that Luisa made and if anybody wants to have this handout, all you have to do is ask.

FYI: Pending Grant Proposal to NFR

Dear All,

Firstly for information: I have applied for a grant from the Norwegian Research Council for a project on scalar structure across categories. We will know in November whether we have been successful and if we can advertise a postdoc position. Luisa Marti and I are going to be organising a reading group here on the stuff related to this grant this term, also starting with prepositions. More on the schedule as it develops. I wonder if it might be a good idea to coordinate reading schedules so that members of the blog are reading the same things? Oh, and we should get Luisa onto this blog. How do I do that, Øystein?

In the mean time, I append the introduction to the grant here for your interest. Any comments and reactions most welcome. If you wish to see the whole 12 page document, just send me an email and I can send you an attachment.

Gillian

Scalar Structure Across Syntactic Categories
1. Introduction and Summary
In our first grammar classes we are often told that ‘A Noun is a name for a person, place, animal or thing’, or, ‘A Verb is a ‘doing’ word.’, with similar notional definitions for the other basic lexical categories (Adjective and Preposition). Although modern syntactic theory has not managed to improve very much on traditional notional definitions of the major parts of speech, practically every theory of grammar makes use of distinct syntactic categories, justified on the basis of formal morphological and distributional criteria. This is particularly true of the traditional ‘lexical’ categories Noun, Verb, Adjective, and to a lesser extent Preposition which have been argued to exist in most, if not all of the world’s languages (see Baker 2003 for a recent study). While it has long been acknowledged, since the advent of X-bar syntax that there are structural parallels across categories (Chomsky 1970, Abney 1987), it has often been implicitly assumed that the major categories are semantically different. In this project, I propose to look critically at this assumption and examine phenomena where the different syntactic categories share important semantic properties and internal relations. In particular, I will examine the
phenomenon of scalar or part-whole structure within the reference types of the major syntactic categories. I seek to demonstrate that all categories show pervasive semantic properties and internal relations. In particular, I will examine the phenomenon of scalar or part-whole structure within the reference types of the major syntactic categories. I seek to demonstrate that all categories show pervasive and linguistically relevant denotational distinctions with respect to reference, systematically contrasting homogenous undifferentiated reference with extended, or linearly ordered scalar structures. The idea here is that the abstract structuring principles of reference are the same across categories, even though the basic ontological domains are different--- Substance (N) vs. Space (P) vs. Property (A) vs. Time (V).

The project will study a three basic domains where part-structure or scalar structure has been shown to be linguistically relevant, and where elements of different syntactic categories interact: (i) explicit measure constructions internal to V, N, A and P; (ii) interactions between temporal and nominal structure in Verb-Object collocations; and (iii) interactions with V and secondary predicates of the A and P type in resultatives. Detailed data of semantic judgements and entailments will be gathered for three main languages: English (representing the Germanic group), Spanish (representing the Romance group where differences within the expression of goal and result have already been noticed), and one non-IndoEuropean language. The first aim of the project is thus empirical and seeks to establish for a group of different languages exactly in which linguistic constructions we find constraints requiring mapping relationships between distinct scalar paths. The second major goal is to formalize a set of semantic composition rules that are sufficiently abstract and flexible to yield parallel results for syntactic categories representing semantically distinct ontological categories (Space, Time, Property and Substance). The final goal of the investigation is to provide a basis for arguing whether the different syntactic categories involve either exactly the same, or merely similar and parallel functional structure within the context of a universal grammar of hierarchically ordered elements. To this end, we will examine the extent to which the morphological and derivational relationships among categories support a common core. The project is built around a project leader and named postgraduate researcher with skills in formal semantic theory and designed to integrate with the more purely syntactic research issues currently being pursued at the Center for Advanced Study in Theoretical Linguistics (CASTL) at the University of Troms\o \ (see sections 3.4 and 4).