LINGUIST List 32.1302
Tue Apr 13 2021
Review: Syntax: Schulte im Walde, Smolka (2020)
Editor for this issue: Jeremy Coburn <jecoburnlinguistlist.org>
Doug Merchant <dochme
The role of constituents in multiword expressions E-mail this message to a friend Discuss this message
Book announced at https://linguistlist.org/issues/31/31-951.html
EDITOR: Sabine Schulte im Walde
EDITOR: Eva Smolka
TITLE: The role of constituents in multiword expressions
SUBTITLE: An interdisciplinary, cross-lingual perspective
SERIES TITLE: Phraseology and Multiword Expressions
PUBLISHER: Language Science Press
REVIEWER: Doug Merchant, University of Georgia
This is the fourth volume in the Phraseology and Multiword Expressions series from Berlin’s Language Science Press, and includes seven research articles spanning psycholinguistic, corpus, and computational investigations into a range of issues surrounding the production, comprehension, representation, and interpretation of complex expressions. The umbrella term used here is “multiword expressions” (MWEs), a term that the editors define extensionally to include particle verbs, noun-noun compounds, and phrasal idioms; the “constituents” of the title are not necessarily syntactic constituents, but rather refer to the various parts of a complex expression.
The initial chapter, by Sylvia Springorum and Sabine Schulte im Walde, considers German particle verbs (PVs, e.g., “anhäufen,” ‘to pile up’). Such verbs are sometimes considered to be noncompositional, with fully idiosyncratic semantics, and even if the particle does contribute anything to the semantics of the verb, the nature of the contribution clearly varies, such that “an” (e.g.) bears a cumulative meaning in one verb (“anhäufen”), a partitive meaning in another (“anbeiβen,” ‘to take a bite’), and a topological meaning in another (“anbinden,” ‘to tie to’). The question the authors address here is whether any conceptual consistency can be identified across these combinations.
Assuming that particles, like prepositions, function in part to spatially ground concepts in what Lakoff (1987) has referred to as “image schemas,” the authors report here on a novel experiment in which participants matched German PVs with directional pictographs referred to as “concept images” consisting of boxes with directional arrows. For example, for the particle “an,” the authors hypothesized that participants would choose images with horizontal arrows pointing to the left and/or right, proposing that the “direction of human sight – with a neutral head position which is horizontal by default – determines this conceptual direction” (p. 11). The results are illustrated descriptively with heat maps based on proportions of responses (no inferential statistics were apparently attempted), and indeed, the authors found for “an-” verbs that more participants chose horizontal arrows pointing to the right than any other pictograph. The authors therefore conclude that directional concepts play a role in semantic composition in particle verbs.
The next article in this volume, by Sandro Pezzelle and Marco Marelli, looks at the classification of compounds, and whether classifications predicted by purely semantic variables in a computational model concord with theoretically-predicted classifications. The framework they use for the latter, taken from Bisetto and Scalise (2005), includes Subordinate compounds such as “doghouse,” which can be rephrased syntactically as “house of the dog”; Attributive compounds such as “swordfish,” where the first constituent acts similarly to an attributive adjective; and finally Coordinate compounds such as “comedy-drama,” whose constituents the authors claim are linked by “an underlying conjunction” (p. 34). This tripartite linguistic framework, the authors note, is similar to that proposed in the psychological literature on conceptual combination. Considering the novel compound “zebra-horse,” for example, one might read a Relation-linking relationship (e.g., a horse which interacts with zebras in some manner), a Property-mapping relationship (e.g., as a horse with some property of zebras’, such as being striped), or a Hybrid/Conjunctive relationship (e.g., an animal which is in its essence both horse and zebra).
This conceptual combination framework is semantic in nature, but some commonalities are apparent with the “syntactic” categories, and so Pezzelle and Marelli set out to test whether semantic variables can predict these classifications, using a computational distributional semantic model (cDSM) in which lexical meanings are encoded geometrically as vectors based on the contexts in which words appear in a corpus. The results (in part) indicated that Coordinate compounds (e.g. “comedy-drama”) are predicted over Subordinate and Attributive compounds by the semantic similarity of the head and its modifier, and that Subordinate compounds (e.g. “doghouse”) are predicted over Attributive by the similarity between the compound and its constituents. The authors conclude that the qualitative, discrete linguistic classifications and continuous, quantitative variation in the meaning of the constituents of a compound are “two sides of the same coin.”
The third article, by Gianina Iordǎchioaia, Lonneke van der Plas, and Glorianna Jagfeld, focuses more narrowly on deverbal compounds (DCs) such as “budget assessment.” Following Grimshaw (1990), the authors argue that deverbal nouns such as “assessment” are ambiguous between an argument structure nominal (ASN) and a result nominal (RN) interpretation. Deverbal nouns with an ASN reading have available what the authors refer to here as “verbal event structure”; this structure “enforces and constrains argument realization in ASNs. . . and its absence in RNs” (p. 63). The contrast can be seen in their (4a-c), reproduced in part here:
(1) a. The examination / exam was on the table.
b. The examination / *exam of the patients took a long time.
c. *The examination of the patients was on the table.
In (1a), we have synonymy between the RN “exam” and the RN reading of “examination,” a synonymy that disappears when the object argument “of the patients” is added in (1b), which results in only the eventive ASN reading being available. In (1c), we see ungrammaticality due to an ASN reading of “examination” enforced by the realization of the argument in a context where only a concrete RN would normally be possible (“on the table”).
Compounds based on deverbal nouns (DCs) may in theory be based on either the ASN or the RN reading of the noun. When it is the former, the authors argue, then the non-head is interpreted as the object of the verb. For example, in “budget assessment,” the interpretation of “budget” as the object of “assess” suggests that “assessment” is unambiguously an ASN, as RNs do not have argument structure. For a DC such as “government assessment,” however, there is not only an ASN-object reading (where the government is being assessed) but also an RN-subject reading (where the government is doing the assessing). Iordǎchioaia et al. claim that the former is the default reading, and that in the second reading, the non-head “behaves just like an adjunct/modifier, since it does not play any role in the compositional makeup of the DC” (p. 65). The authors hypothesize that DCs whose heads are ASNs will realize only objects as non-heads; unlike RNs, they are compositionally constructed. RNs, on the other hand, are lexicalized, and their apparent deverbal derivational morphology is irrelevant; they are in this respect similar to non-derived nouns.
To test this hypothesis, the authors examined DCs for a set of morphosyntactic and semantic properties Grimshaw had identified as specific to ASNs. One of these they term “process-vs-result” (P-R), or whether the head of a DC refers to an ongoing process or its result; this and other features were obtained from native-speaker annotations. Iordǎchioaia et al. then used a machine-learning technique in which they trained a logistic regression classifier to model whether these features predict the relationship between the head and the non-head, in particular whether the non-head was an object or non-object. As it turned out, the greatest predictor was not a morphosyntactic property extracted from the corpus, but a semantic feature introspected by the annotators: “process-vs-result.” The authors conclude that DCs are compositional only to the extent that their heads evince ASN properties; such compounds are also more transparent than those whose heads evince RN properties.
In the fourth article here, Gary Libben considers novel compounds, reporting on the results of a progressive demasking / typing experiment which investigates whether all possible lexical substrings are activated in orthographically ambiguous novel compounds such as “clampeel” (parsable as either CLAMP-EEL or CLAM-PEEL). Libben argues that such compounds cannot be said to “have” any morphological structure when first encountered; rather, structure is imposed by the parser. Libben proposes as a heuristic Fuzzy Forward Lexical Activation (FFLA), wherein initial letters are scanned left-to-right until a known substring is recognized (e.g., CLAM). At this point, a final substring is posited beginning with the next letter (PEEL); if it is also recognized, it’s interpreted, and the procedure resets. On the second parse, the initial letters are again scanned until another known substring is recognized (CLAMP); a final substring is then posited (EEL), and the procedure ends. This heuristic thus generates two disparate representations for the same string (CLAM-PEEL and CLAMP-EEL); in allusion perhaps to the physicists’ phenomenon of quantum superposition, Libben describes this as a “lexical superstate.”
Libben then reports on the results of an experiment in which participants retyped both ambiguous (CLAMPEEL) and nonambiguous (ANKLECOB) compounds presented on a screen, with the typing latencies of each letter recorded. A significant spike in latencies between the two possible letter strings for unambiguous compounds (ANKLE - COB) was apparent, with only moderately longer latencies at the two possible boundaries in ambiguous compounds (CLAM – PEEL; CLAMP - EEL). Libben takes this as evidence for lexical superstates and for FFLA, noting however that the latter may be an adaptation specific to English, as other languages (e.g. German) have orthographic systems which represent compounds differently.
This observation provides an excellent segue into the fifth article, a qualitative corpus study by Inga Hennecke on variability in the preposition in N Prep N constructions in Romance. For example, Spanish “juego de niños” (‘children’s game’) may also be realized as “juego para niños,” apparently without any semantic distinction. A lack of internal variation is widely seen as a diagnostic for compoundhood (and for idiomaticity), and the debate over the significance of the prepositional variation in these compounds therefore stands as a potential proxy for the debate over the boundary between syntax and morphology. Are such phrases the result of productive syntactic processes? Or should they rather be classed as fully lexicalized compounds which admit minor (idiolectal?) variation.
Hennecke looks for a middle path here, arguing that “it is neither necessary nor possible to draw a clear distinction between syntactic constructions and lexical constructions” (p. 135). Working within the framework of construction grammar (Goldberg, 1995), and in particular Booij’s (2015) construction morphology, Hennecke argues that the prepositions in N Prep N constructions are not semantically opaque linking elements but rather have “a specific semantic value determined by the semantic functions of the nominal constituents” (p. 142). She supports this argument through a crosslinguistic analysis of the semantics of the nominal constituents in a range of such constructions in French, Spanish and Portuguese. In considering (e.g.) the contrast between “de” and French “pour” / Spanish and Portuguese “para,” Hennecke concludes that in French, “pour” is only possible where the second noun is in some sense a user or beneficiary, or where it specifies the purpose of the first noun: “collier de chien,” “collier pour chien” (‘collar of a dog’, ‘collar for a dog’; Hennecke therefore posits an abstract semantic template which registers this relationship and governs this construction. A similar situation is apparent for Spanish “de”/”para”; however, in Portuguese, variation between “de” and ”para” is possible for a larger number of semantic relations between the two nouns, such that (e.g.) the first noun may serve as an instrument for the second noun: “creme de/para mãos” (lit. ‘cream of/for hands’, id. ‘hand cream’).
The penultimate piece in this volume, by Christina L. Gagné, Thomas L. Spalding, J. Claire Burry, and Jessica Tellis Adams, reports on an experiment on production choices between bare nouns (“dog”), noun phrases with prenominal modifiers (“blue dog”), which the authors refer to as “modifier-noun phrases,” and noun phrases with post-nominal information (“a dog that was blue”), or “full phrases.” The question here is twofold: does the distinctiveness of modifying information about an entity in a narrative influence whether that information is reproduced when a speaker subsequently refers to that entity? And does the syntactic form in which a nominal is modified in a narrative influence the syntactic form in which a speaker conveys that information? The first question concerns the manner in which a speaker marks conceptual distinctions among member of a category; for instance, a blue dog is more conceptually distinct from the larger category of dogs than is a brown dog. The second question, of course, concerns the phenomenon of syntactic priming, i.e., whether exposure to given syntactic structures in some input impacts the structure chosen as output.
The experiment itself is concisely constructed, with four narrative conditions: nondistinctive / modifier-noun phrases (“brown dog”), non-distinctive / full phrases (“dog that was brown”), distinctive / modifier-noun phrases (“blue dog”), and distinctive / full phrases (“dog that was blue”). After reading narratives which included target information, participants were asked to answer questions about the story (by typing) which necessitated reference to the target phrase. Both the distinctiveness of the information and the syntactic form were significant in the analysis; perhaps to no one’s surprise, participants were more likely to include distinctive information, and more likely to answer with bare nouns for nondistinctive information. Interestingly, they were also significantly more likely to include modifiers when the information was presented as a modifier-noun phrase than when it was presented post-nominally. Full phrases in the input were more likely to induce full phrases in the output; however, modifier-noun phrases were still preferentially produced even in this condition. Gagné et al. conclude that conceptual distinctiveness influences the choice to include modifying information, and that a general preference exists for modifier-noun phrases over full phrases when it comes to including that information.
Finally, a study by Eva Smolka and Carsten Eulitz considers the phrasal idiom (e.g., “reach for the stars”), the paradigm of a complex expression. Although the subtitle of the article references “idiom processing,” Smolka and Eulitz used an offline task to measure speakers’ judgements on meaning relatedness between modified German idioms like “nach den Sternen greifen” (lit. ‘reach for the stars’, idiomatically same as English) and paraphrases of the idioms’ meanings. Their intention was to examine whether a figurative reading is still available when one of the idiom’s parts has been modified (“reach for the planets”), and thus whether idioms are fully “semantically fixed.” (What is meant by “semantically fixed” is not explicitly defined; one wonders whether “phonologically-” or “orthographically-fixed” might not be more appropriate, as the semantics of the idioms were not modified, but merely the form.)
Three experiments were conducted: one in which the nouns varied, one in which the verbs varied, and one in which the preposition varied. In the first experiment, the authors replaced the noun with either a semantically related noun (“Planeten,” ‘planets’) or an unrelated noun (“Bonbons,” ‘sweets’). In the second, the verb was replaced with either an associated verb (“gelangt,” ‘grasped’) or an unrelated verb (“gefragt,” ‘asked’). The third experiment, in which the prepositional was varied, looked slightly different, in that either a different preposition appeared (“zu,” ‘to’) or a prepositional phrase with the same preposition but a different nominal appeared (“nach den Bonbons,” ‘for the sweets’). Trials consisted of pairing a paraphrase of the meaning of the idiom with a sentence containing either the unaltered or the altered idiom; participants rated on a seven-point scale the similarity in meaning of the two sentences.
As one might expect, the sentences with a slightly altered constituent were rated as closer in meaning to the paraphrases than were the sentences with an unrelated constituent; this effect was larger in the first experiment, where the verb was switched out, than in the conditions where the noun or preposition was replaced. Smolka and Eulitz conclude that the figurative meanings of idioms are “activated” even when part of the idiom has been supplanted, and that modified constituents “may contribute to the generation of the figurative meaning of the idiom” (p. 197). They take this as evidence for views like that of Hamblin & Gibbs Jr. (1999), in which the semantics of a verb in its literal sense may (synchronically) constrain the semantics of the idiom in which the same form appears (this is also argued for in Marantz, 1997 and McGinnis, 2002; the opposite position is articulated in Merchant, 2019). The authors here further argue that their results tell against “any type of model on idiom comprehension or production that assumes some kind of fixed lexical entry of the idiomatic constituents that generate the figurative meaning” (p. 199), such as e.g. the “superlemma” theory of Sprenger et al., 2006. As the experiments reported here did not investigate online comprehension but rather offline meaning comparisons, this claim especially will probably strike most psycholinguists as a bit too far-reaching.
This volume is interdisciplinary by design, and so cannot be cohesive in a strong sense. However, the articles included all address theoretical questions with some sort of empirical study, whether corpus-based, experimental, or computational, and thus together give a good sense of the range of approaches researchers are bringing to bear on understanding multiword expressions.
More importantly, however, is the cohesion in subject matter, and that is the chief contribution of this volume as a whole. Rarely are particle verbs and phrasal idioms brought under the same roof, despite having some important properties in common, and anyone who reads this book from cover to cover will doubtlessly come away with some important insights. Of course, research articles are not normally consumed in this manner, so I will briefly evaluate the articles included here in turn.
The first article, by Springorum and Schulte im Walde, consisted of an experimental study on German particle verbs; the authors sought to determine whether speakers would reliably match directional pictographs to particles regardless of the verb with which they combine. The chief flaw of this study is that despite reams of quantitative data, no inferential statistics were apparently attempted. The lack of so much as an ANOVA is puzzling (especially given that other articles in this volume use state-of-the-art statistical techniques); any experimentalist worth their salt will therefore likely look on the authors’ conclusions skeptically. How are we to know that the apparent agreement of participants on the pictographs is not due to random chance?
On a more minor note, there are a few occasions where the authors cite evidence only indirectly which might help their case if cited directly: “[A]ccording to Gibbs Jr. & Colston (1995) there is good evidence that both spatial and visual representations exist for mental imagery” (pp. 5-6). Especially since the general theoretical framework referenced here (holding that conceptual structures exist independent of language) is somewhat controversial among mainstream linguists (see, e.g., Keysar & Bly, 1999), such evidence should probably have been described in some detail in this section.
The second article here, by Pezzelle and Marelli, looked at whether “syntax-based” classifications of compounds as Subordinate, Attributive, or Coordinate can be identified via continuous semantic properties in a computational distributed semantics model. The study is well-done, and the authors wise to note that “discrete and continuous approaches are two faces of the same coin,” giving neither primacy. A small note would be the possibly misleading use of the terms “syntax-based” and “syntactic” to describe the categories; there’s very little of interest to the average syntactician here, nor are the categories justified in terms of syntactic data (e.g., example sentences).
The third article, by Iordǎchioaia, van der Plas, and Jagfeld, looks at deverbal compounds with ASN readings and whether a set of semantic and morphosyntactic properties theoretically proposed to be specific to ASNs can predict the relationship between the head of a compound and its non-head modifier. The theoretical issues involved go back to Chomsky’s (1970) “Remarks on nominalization,” and arguments for and against Chomsky’s “lexicalist” position continue to this day (e.g., Bruening, 2018). The authors here profess to not be taking sides in this debate, but their conclusion that ASNs are indeed compositional nevertheless seems to argue against the lexicalist position, so some greater discussion of the theoretical issues would have been useful. It might also be mentioned that a number of their grammaticality judgements are likely not universal, with counterexamples readily available. For example, the authors note that “The examination of the patients was on the table” (their (4c)) is unacceptable, but “The assessment of the patients was on the table” is quite felicitous to this reviewer’s ears. (I also find no contrast whatsoever between (9a) and (9b).)
Libben’s article on novel compounds and his Fuzzy Forward Lexical Activation model will be of great use to anyone who works on orthographic processing, although whether an analogical process in the domain of speech processing might be found is uncertain. It will also be of interest to anyone who works on morphology generally, and likely anyone concerned with how text is processed in advertising. It’s also a clear and concise read, with a well-designed and novel methodology that any experimentalist will appreciate.
Inga Hennecke’s qualitative corpus study of left-headed N Prep N compounds in Romance is a useful addition to the literature, and will provide some much-needed perspective for anyone who usually considers only right-headed Germanic nominal compounds; indeed, anyone interested in the boundary between syntax and morphology would do well to consider such facts. Those who are already aficionados of construction grammar will likely be more convinced of the analysis than those who are not as familiar with such approaches; the section which introduces construction morphology is fairly brief and is probably not easily apprehensible for novices.
The production choice experiment reported on in in Gagné et al. is well-designed and perspicuously laid out, but feels a bit out of place in this volume in that the multiword expressions considered here are not complex fixed expressions or in any manner noncompositional (this is technically true of Libben’s novel compounds, too, but many fixed expressions begin their lives as novel compounds, so his study seems relevant). It is an important experiment, however, in a basic-science sort of way; the fact that only distinctive information is likely to be conveyed in a summary of that information might seem obvious to some, but it is important to demonstrate the obvious experimentally so that one might then felicitously probe the not-so-obvious.
Finally, the experimental study by Smolka and Eulitz (the only one in the volume that is concerned primarily with figurative language) is another well-designed study which demonstrates the obvious, viz., that the semantics of an idiom are more likely to be considered when one of its constituents is replaced by a word whose meaning is similar to the word it replaces than when the word is replaced by an unrelated word, so that “reach for the planets” is judged as more similar in meaning to ‘aspire to unobtainable goals’ than is “reach for the sweets.” However, as I note above, the conclusions reached by the authors do not seem to be warranted on the basis of this study. Collecting offline judgements of meaning similarity, in which participants have an unbounded amount of time to compare sentences (the experiments here were conducted by email), tells us next to nothing about idiom processing or semantic activation.
The chief conclusion that can be drawn from the data reported here, it seems to me, is merely that the participants were familiar enough with the idioms involved to be able interpret a phrase like “reach for the planets” as ‘aspire to something unobtainable’. Note that “reach for the planets” does not have a plausible literal reading, and so the only available interpretation is a metaphorical one, which “alludes to an idiom without actually containing one,” as Svenonius (2005:228) puts it. In short, it is a metalinguistic reference to an idiom, made possible by introspection. What would be interesting would be an online study of this phenomenon, wherein a speaker utters a modified version of an idiom (as might happen by accident) and a listener is forced to recover the meaning of the original idiom under ecologically valid conditions. Such a study might justify the sorts of conclusions the authors reach here.
Bisetto, A., & Scalise, S. 2005. The classification of compounds. Lingue e linguaggio 4(2). 319-32.
Booij, G. 2015. Word-formation in construction grammar. In O.P. Müller, I. Ohnheiser, S. Olsen & F.
Rainer (eds.), Crossdisciplinary issues in compounding (Current issues in linguistic theory 311), 93-109. Amsterdam, John Benjamins.
Bruening, B. 2018. The lexicalist hypothesis: Both wrong and superfluous. Language 94(1). 1-42.
Chomsky, N. 1967 . Remarks on nominalization. In Roderick A. Jacobs & Peter S. Rosenbaum (eds.), Readings in English transformational grammar. Waltham, MA: Ginn.
Goldberg, A.E. 1995. Constructions: A construction grammar approach to argument structure. Chicago/London: The University of Chicago Press.
Grimshaw, J. 1990. Argument structure. Cambridge, MA: MIT Press.
Hamblin, J.L. & Gibbs Jr., R.W. 1999. Why you can’t kick the bucket as you slowly die: Verbs in idiom comprehension. Journal of Psycholinguistic Research 28(1). 25-39.
Keysar, Boaz & Bly, Bridget M. 1999. Swimming against the current: Do idioms reflect conceptual structure? Journal of Pragmatics 31. 1559-78.
Lakoff, G. 1987. Women, fire, and dangerous things: What categories reveal about the mind. Chicago: The University of Chicago Press.
Marantz, A. 1997. No escape from syntax: Don’t try morphological analysis in the privacy of your own lexicon. University of Pennsylvania Working Papers in Linguistics 4(2):201-25.
McGinnis, M. 2002. On the systematic aspect of idioms. Linguistic Inquiry 33(4). 665‐72.
Merchant, D.C. 2019. Idioms at the interface(s): Towards a psycholinguistically grounded model of sentence generation. Athens, GA: University of Georgia dissertation.
Sprenger, S.A., W.J.M. Levelt & G. Kempen. 2006. Lexical access during the production of idiomatic phrases. Journal of Memory and Language. 54(2). 161-84.
Svenonius, P. 2005. Extending the Extension Condition to discontinuous idioms. In Pierre Pica, Johan Rooryck & Jeroen van Craenenbroeck (eds.), Linguistic Variation Yearbook 5. 227-63.
ABOUT THE REVIEWER
Doug Merchant holds a PhD in Linguistics from the University of Georgia (UGA) and a BA in same from the University of Arizona. His dissertation examined what the syntactic and semantic properties of phrasal idioms reveal about the lexical interface(s). He has published also on the question of the directionality of derivations, with a focus on theory-external arguments for top-down structure building, and is currently writing up the results of an experimental study on empty category processing in Brazilian Portuguese. He was most recently a visiting researcher at UGA, and has also taught syntax at San Diego State University.
Page Updated: 13-Apr-2021