ON THE ARGUMENT-ADJUNCT DISTINCTION IN THE POLISH SEMANTIC SYNTAX TRADITION1

The aim of this paper is to examine the understanding of the Argument-Adjunct Distinction within the Polish Semantic Syntax (SS) tradition, associated with the name of Stanisław Karolak and presented in the nominally syntactic volume of the Grammar of contemporary Polish (Pol. Gramatyka współczesnego języka polskiego; Topolińska, 1984), especially in Karolak (1984) and Grochowski (1984), as well as in later work. Section 1 reviews the three approaches to determining the number and kind of arguments of a given predicate, as discussed in Karolak (1984), concentrating on the one that is endorsed there. Then, Sections 2–3 show that the key notions used in this approach have not been – and probably cannot be – made operational. Moreover, Section 4 briefly reviews some more recent Semantic Syntax work and shows that this lack of operational AAD in Karolak (1984) extends to the SS programme at large. Finally, Section 5 concludes that this deficiency, while common in linguistic theories, is particularly troublesome in the case of SS, which is founded on the notion of predicate-argument structure.

1 Three approaches to AAD Determining a predicate's argument structure is of paramount importance to Semantic Syntax.As the only "steadfast procedure for the adequate identification of the number of arguments in a predicate-argument structure", Karolak (1984, p. 54) mentions the decomposition of lexical senses, best known from the work of Anna Wierzbicka, and applied to Polish, for example, in On the argument-adjunct distinction in the Polish Semantic Syntax tradition Grochowski (1980).However, Karolak (1984, p. 55) also rightly -if somewhat euphemisticallynotes that this procedure "is often treated as not being entirely free of aspects of subjectivity".
A more objective method would rely on the notion of syntactic obligatoriness, i.e., the mandatory textual co-occurrence of argument expressions, corresponding to semantic arguments, with a textual realisation of a given semantic predicate.But Karolak (1984, pp. 55-56) also notes that while obligatory expressions usually denote arguments, optional expressions are not necessarily adjuncts.This echoes the widespread linguistic intuition, also applicable to English, expressed in the following quotes from syntactic textbooks (Borsley, 1991;Tallerman, 2005), a recent handbook article on arguments and adjuncts (Ackema, 2015), and a recent monograph on argument realisation (Williams, 2015): (1) "[C]omplements tend to be obligatory, whereas adjuncts are always optional."(Borsley, 1991, p. 62) (2) "Adjuncts are always optional, whereas complements are frequently obligatory."(Tallerman, 2005, p. 98) (3) "[I]n contrast to arguments, adjuncts are never obligatory."(Ackema, 2015, p. 264) (4) "[I]f removing a dependent results in unacceptability, this is good initial evidence that the omitted dependent is an argument. . .The converse is not on equal footing, however.If a dependent can acceptably be omitted, it need not be an adjunct. . ." (Williams, 2015, p. 68) For this reason, Karolak (1984, p. 56) settles on a third approach to ascertaining a predicate's argument structure, one that invokes the notion of semantic obligatoriness: if removing an expression results in semantic ill-formedness or alters the meaning of the textual realisation of a predicate, then the expression corresponds to the predicate's argument.Note that this is again, as in the case of syntactic obligatoriness, a unidirectional implication: semantic obligatoriness implies argumenthood, not the other way around.However, in the case of semantically optional expressions, exactly these are adjuncts, which represent a separate predicate-argument structure.Hence, this third approach may precisely bifurcate a predicate's dependents into arguments and adjuncts, but only to the extent that the two crucial notions it relies on, semantic obligatoriness and separate predicate-argument structure, may be made operational.

Semantic obligatoriness
2.1 Karolak (1984) The following examples are among those adduced in Karolak (1984, pp. 56-57)  Only some of the italicised expressions are considered to be semantically obligatory, whereas the other are considered to be semantically optional.However, the judgements of native speakers of Polish vary here.(At this point, before reading on, Polish readers may want to form their own opinion on these examples.)According to Karolak, the italicised phrases are semantically optional in ( 6), (8), and (9), and semantically obligatory in the other three examples.But, to the best of our knowledge, it has never been demonstrated that the notion of semantic obligatoriness is intersubjectively stable, i.e., that linguists agree on the status of examples such as (5)-( 9) above, with the italicised expressions omitted.
For the purposes of this article, an experiment was performed, in which ten subjects2 judged the semantic obligatoriness of the italicised phrases in the above examples.The subjects were asked whether, after removing the phrase in question, the sentence became semantically incomplete.They could assign each sentence to one of the following five categories: 1: the sentence definitely becomes semantically incomplete after removing the phrase, 2: the sentence becomes semantically rather incomplete after removing the phrase, 3: difficult to say (subjects were asked to avoid this option), 4: the sentence rather stays semantically complete after removing the phrase, 5: the sentence definitely stays semantically complete after removing the phrase.
Most variation in judgements is observed in the case of ( 10).This may be attributable to the fact that the reduced version, Jan cię prosi, represents the complete meaning 'Jan is inviting you On the argument-adjunct distinction in the Polish Semantic Syntax tradition (to enter)'.Hence, the judgements reflect the granularity of word sense distinctions assumed by the subjects (see Kilgarriff, 1997 on this issue): if 'invite to enter' is understood as a specialised sense of 'ask', then 5 (or 4) should be assigned to (10); if, however, it is understood as a sense distinct from 'ask', the assignment of 1 (or 2) is justified.
Another example used in this experiment is (12), from Grochowski (1984, p. 256), where the italicised phrase is considered to express an argument: (12) Maria Maria wróciła returned znad from morza.sea 'Maria returned from the seaside.' The ten judgements are: (13) ex s1 s2 s3 s4 s5 s6 s7 s8 s9 s10 av stdev (12) 5 5 5 5 2 5 4 1 2 2 3.6 1.6 While six subjects considered the ablative expression to be semantically optional, four pronounced it semantically obligatory.Such variations in judgements on semantic completeness, and the stark differences between the theory-laden judgements in Karolak (1984) and the pre-theoretical judgements reported here, undermine the operational status of the intuition of semantic obligatoriness, as understood in Semantic Syntax.

Panevová (1974)
A different test for semantic obligatoriness is the dialogue test proposed in Panevová (1974) Such dialogues are concerned with expressions which are not syntactically obligatory.In the case of ( 14), the amount that was paid does not have to be expressed on the surface when the verb buy is used, and the naturalness of the dialogue also shows that it is not semantically obligatory.
On the other hand, the unexpressed merchandise in ( 15) is semantically obligatory to the extent that the reply of A does not sound natural (as indicated by # ): how could she have said John paid Harry five dollars if she did not know what he paid for?Apparently, such semantically obligatory arguments must be known to the speaker, or perhaps contextually provided, in order to be omissible on the surface.One of the standard applications of the dialogue test, cited in various works by Jarmila Panevová and her colleagues (also in Panevová, 1974, pp. 16-18, is to investigate the nature of ablative and adlative dependents of various verbs of motion; in the case of verbs such as arrive or return, an ablative expression seems to be semantically optional, and an adlative expression seems to be semantically obligatory.The same contrast is observed in the case of Polish, as dialogue (17) sounds much less natural than ( 16): (16) A: Maria właśnie wróciła!B: Skąd?A: Nie mam pojęcia.
A: Maria has just returned!B: Where from?A: I have no idea.
A: Maria has just returned!B: Where to?A: # I have no idea.Note that this directly contradicts the view that ablative phrases in examples such as (12) above are semantically obligatory.
As discussed at length in Przepiórkowski (2016a), although the dialogue test seems to give a clearer indication of semantic obligatoriness than the intuition about out-of-context examples assumed in the Semantic Syntax approach, its operational value is also limited.In particular, in the minimalism vs. contextualism debate (within the philosophy of language), an attempt has been made to apply this test to decide on the status of the syntactically optional locative expression in sentences such as It is raining (somewhere), but the results are rather inconclusive.While Recanati (2007Recanati ( , 2010) ) constructed elaborate contexts in which the dialogue (18) sounds natural, his opponents, cited in Recanati (2010, pp. 115-117), showed that with sufficiently sophisticated contexts, even direct objects of verbs such as notice and finish, generally assumed to be semantically obligatory, may come out as semantically optional, as in (19).

Partial conclusion
Given that no intersubjective procedure for determining semantic obligatoriness has been proposed within Semantic Syntax, and no such procedure is in sight in other approaches, one must conclude that the first component of the method adopted in SS to distinguish adjuncts from arguments is not operational.This alone justifies the conclusion that there is no operational understanding of the Argument-Adjunct Distinction in Semantic Syntax.
3 Separate predicate-argument structure Let us ignore the result of the previous section and assume that it is possible to distinguish between semantically obligatory and semantically optional expressions.In Semantic Syntax, the former correspond to arguments, but in the case of the latter an additional criterion decides whether they correspond to arguments or to adjuncts, namely, the criterion of constituting a separate predicate-argument structure (henceforth, SPAS).Karolak (1984, p. 56) does not provide a method of determining whether or not a given expression constitutes a SPAS, referring the reader to Bogusławski (1974) instead.
Unfortunately, very little is said there within the two paragraphs devoted to this matter.Bogusławski (1974, p. 45) juxtaposes two constructions: pisać czymś 'write with something' and jechać po coś 'go to get something', as in the following two examples: (20) Jan Jan pisze writes list letter piórem.pen 'Jan is writing a letter with a pen.' 3 One such context, attributed to Sam Wheeler III, is this: Therapists are monitoring a large group of patients suffering from Fred's Syndrome, a pathological tendency to start projects and never complete them.A new drug, Completin, is being tested.Patients are monitored by undergraduate students, who push a button every time a patient finishes a project.'Patient #271 has finished,' says the researcher, looking at the console.'John has finished' can mean 'John has finished something or other'.
On the argument-adjunct distinction in the Polish Semantic Syntax tradition (21) Jan Jan jedzie goes po for gazetę.newspaper 'Jan is going to get a newspaper.' The expressions piórem 'with a pen' in (20) and po gazetę 'to get a newspaper' in (21) are both semantically optional; the two sentences are semantically complete without them.The purported difference is that the meaning of one of these expressions "does not involve the superordinate scheme" introduced by the main predicate, i.e., it constitutes a SPAS, and the meaning of the other expression does involve the meaning of the main verb, i.e., it does not constitute a SPAS.(Again, the Polish reader is invited to form an opinion at this stage about which expression is supposed to also involve the superordinate meaning, and which does not.) According to Bogusławski (1974, p. 45), the expression po gazetę 'to get a newspaper' meets the SPAS condition, as the meaning of po gazetę, namely, "X is going to L because X wants a newspaper", is independent of Jan jedzie, which may be translated as 'Jan is going (by car, by bus, by train, etc.)' or 'Jan is driving ' and, according to Bogusławski (1974, p. 45)" "does not involve any direction or purpose. . .there is only a temporal-causal relation between the two facts".On the other hand, piórem has the meaning of "X is using a pen", which -according to Bogusławski -is "hardly a full sentence scheme without some in writing, for writing, or the like".
In the sentence just cited, an implicit reference to semantic obligatoriness seems to be made, which has already been shown in the previous section to be problematic.Regardless of this potential problem, this method clearly refers to the procedure of decomposing lexical senses, which is rightly -even if somewhat hesitantly -rejected by Karolak (1984, p. 55) as subjective.This subjectivity is evident in the explications of the two expressions discussed by Bogusławski (1974, p. 45).Just as po gazetę in ( 21) is explicated as "X is going to L because X wants a newspaper", with "going" referring here to the same event as the subordinate verb jedzie 'going (by car, etc.)', also piórem in (20) could be explicated as "X is using the pen in order do something", with "do something" referring to the same event as pisze list 'write a letter'.
Obviously, given the terseness of the presentation of the procedure for recognising a separate predicate-argument structure in Bogusławski (1974) (and, hence, also here), it is very well possible that the current attempt to invalidate this procedure misses some important point implicit in that presentation.But given that the predicate-argument structure is a fundamental notion of the Semantic Syntax approach, and that being able to recognise a SPAS is crucial for the procedure of determining a predicate's arguments, it is surprising that, within the 200-page presentation of Karolak (1984), only a quick reference to Bogusławski (1974) is made, where only a brief passage (comprising of about half a page) is devoted to this issue.
Perhaps this lack of clear elucidation of the notion of separate predicate-argument structure is not an accident, after all, but rather a reflection of a principled impossibility to decide whether a given expression constitutes a SPAS.Consider, for example, the simple sentence John stutters, and assume the simplistic semantic representation of the verb as λx.stutter (x) (ignoring tense, aspect, a possible event variable, etc.).Now, if John is represented as a constant, j, i.e., definitely not as a SPAS, the lambda expression for the verb may be applied to this constant and, after beta-conversion, render the expected representation of this sentence, i.e., stutter (j).However, the representation of John may be type-shifted (Partee, 1986) to λP.P (j), i.e., to a predicate-argument structure of sorts.This lambda expression may now be applied to the representation of the verb and, after a couple of beta-conversions, render the same representation of this sentence.So does John constitute a SPAS, or not?Well, this depends on the underlying semantic theory.While in this case j is a simpler representation than the unnecessarily type-shifted λP.P (j), and so it might seem that the obvious answer is that it is not a SPAS, the standard representation of more complex nominal phrases -those expressing generalised quantifiers -is more similar to the shifted representation, i.e., they are often treated as functors which take verbal arguments.
In summary, the second component of the method adopted in SS to distinguish adjuncts from arguments, based on the notion of a separate predicate-argument structure, is just as non-In later work, for example in a 2002 monograph, Karolak maintains the views discussed above, mentioning again (on p. 99) lexical decomposition of senses as the most principled way of determining a predicate's arguments, and noting again that this procedure is generally considered to be subjective.Just as in Karolak (1984), syntactic obligatoriness is also rejected in Karolak (2002), and the procedure relying on the notions of semantic obligatoriness and separate predicateargument structure (with a reference to Bogusławski (1974) again) is adopted.As is also clear from the guidelines to the project Składnia porównawcza języków słowiańskich drugiej połowy XX w. 'The comparative syntax of Slavonic languages of the 2nd half of the 20th century', Karolak was very well aware of the difficulties related to AAD -he explicitly refers there to the role of the subjective intuition (intuicja autorska; Karolak, 2004, p. 3) in the process of decomposing senses. 5part from the work of Stanisław Karolak, the only other specific discussion of AAD within the Semantic Syntax approach that we are aware of is that of Kiklewicz (2007Kiklewicz ( , 2010)).Four tests for AAD are considered there: syntactic obligatoriness (considered and rejected in Karolak, 1984Karolak, , 2002)), semantic obligatoriness (as adopted in Karolak, 1984Karolak, , 2002)), the dialogue test discussed in Section 2.2 above, and a transformation criterion, according to which adjuncts -but not arguments -of a given predicate may be paraphrased within a subordinate clause.This is illustrated by the following examples, purportedly showing that córce 'daughter' in (22) (Steinitz, 1969;Helbig, & Schenkel, 1973) and forcefully rejected by another German valency researcher, Heinz Vater (1978a, pp. 28-29). 6To the best of our knowledge, this proposed test has not been resurrected ever since and does not feature in contemporary discussions of AAD, so we will also ignore it here.
As to the other three criteria, it is clear that they are pairwise incompatible.Obviously, the syntactic criterion is at odds with either the semantic completeness approach of Karolak (1984), 2002 or the dialogue test of Panevová (1974) -the latter two were explicitly designed to deal with syntactically optional expressions.Additionally, these two semantically-oriented criteria give different results in some cases, as already shown above.To give a different example, in (7) cited On the argument-adjunct distinction in the Polish Semantic Syntax tradition above, which involves the verb gardzić 'despise', the phrase za jego uległość wobec władzy 'for his submissiveness towards the authorities' was judged as semantically obligatory in Karolak (1984, p. 57), while it is clearly semantically optional in the sense of Panevová (1974), as demonstrated by the naturalness of the following dialogue: (24) A: Jan gardzi Piotrem.
A: Jan despises Piotr.B: What for?A: I have no idea.
In fact, on the assumption that the Polish Semantic Syntax tradition and the Praguian Functional Generative Description refer to approximately the same concept of arguments and adjuncts, the two criteria cannot give the same results in principle: the semantic criterion of Karolak (1984Karolak ( , 2002) ) is understood as a complete procedure of distinguishing arguments from adjuncts, while in the Praguian theory the dialogue test of Panevová (1974) is only a part of the complete procedure, which also involves the more syntactically oriented criteria of specificity and iterability.
In summary, it is clear that the criteria discussed in Kiklewicz (2007Kiklewicz ( , 2010)), to the extent that they are operational at all, do not work in unison.Unfortunately, the paper does not explicitly discuss the differences between their results, nor does it contain any recommendations on how to resolve them.Hence, it must be concluded that the Semantic Syntax approach as a whole does not offer an operational understanding of the Argument-Adjunct Distinction.

Conclusion
The Semantic Syntax approach is not unique in lacking an operational definition of AAD; in fact, this seems to be the norm.As discussed in Vater (1978aVater ( , 1978b)), a coherent view is already absent in Tesnière (1959Tesnière ( , 2015)), the founding work on valency.Perhaps surprisingly, after 60 years of intensive research on this topic, we do not seem to be any closer to a clear understanding of this purported dichotomy.While a large number of tests have been proposed, their lifespan is usually short.Moreover, as noted in Tutunjian and Boland (2008, p. 633), "[t]he sheer number of these tests underlines the fact that no single test is entirely satisfactory".So an idea must be entertained that there is no single fundamental Argument-Adjunct Distinction, but rather a continuum of dependents which may be split into different classes according to a number of criteria.
This conclusion may be incompatible with various linguistic theories to various degrees.For example, in the case of Lexical Functional Grammar (Bresnan, 1982;Dalrymple, 2001;Bresnan, Asudeh, Toivonen, & Wechsler, 2015), a theory encompassing not only syntax and semantics, but also other linguistic levels, it seems that relatively conservative modifications to the current theoretical assumptions are needed to make it compatible with the view eschewing AAD -this is argued in Przepiórkowski 2016b.Also, AAD might not be crucial for Functional Generative Description (Sgall et al., 1986) -this claim is made in Przepiórkowski (2016a) (and contested in Panevová, 2016).However, the Semantic Syntax approach is much more vulnerable in this respect, as much of the work carried out within it concentrates on the description of predicateargument structures, especially on determining the number and possible realisations of arguments of particular predicates.Rather than being a general theory of syntax (or semantics), SS covers a similar territory to those modules of much broader contemporary theories that deal with the socalled Argument Structure (see, e.g., Grimshaw, 1990;Jackendoff, 1990;Goldberg, 1995;Wechsler, 1995;Alsina, 1996;Davis, 2001;Levin and Rappaport Hovav, 2005;Babby, 2009;Williams, 2015 -to refer to just a few of the many monographs on the topic published during the last quarter of a century, within diverse linguistic theories), but it has been developed in splendid isolation and, as a result, lacks the flexibility and formal devices of these more contemporary theories.Hence, the absence of an operational distinction between arguments and adjuncts, and the possibility On the argument-adjunct distinction in the Polish Semantic Syntax tradition While a 1985 paper by the Belorussian linguist V. V. Martynov is cited in this context, a similar test was actually proposed much earlier by German valency theorists