Ch1: Introducing Construction Grammar
Traditionally, linguistic knowledge is thought of as having a lexicon and grammar component (the dictionary-and-grammar model), but construction grammar proposes that all linguistic knowledge is different constructions. The change is motivated by idiomatic expressions that are a sort of “appendix” in dictionaries. Yet we can’t represent idioms as fixed expressions because they have slots that can be filled productively. For example, “too big a shock” and “quite useful a lesson” are the same construction, with a predictable meaning and certain constraints on the slots.
Fillmore (1988) was the first to propose this solution to the problem of idioms. Idioms are very common in language and difficult to describe with the dictionary-and-grammar model, so some new mechanism is required. But then this new mechanism can also replace the other parts of the model too.
What exactly is a construction? Goldberg (1995) proposed that a construction is a mapping from form to meaning where some aspect of either the form or meaning is not predictable from previously established constructions. Constructs are specific instances of constructions, for example, “many a day” is a construct and an instance of the MANY-A-NOUN construction. Predictable language can also be constructions if they’re frequent enough and serves as the default for a pragmatic function, eg: “how is your day?”.
Four ways of identifying constructions:
- Look for ways that an expression deviates structurally, eg: “by and large” — unclear what are the POS of each of the words.
- Look for non-compositional meaning, eg: “make waves”.
- Look for idiosyncratic constraints, eg: “asleep” can be used as a predicative, but not attributive adjective.
- Look for collocational preferences, eg: “will” has a slightly different distribution from “going to”.
Ch2: Argument Structure Constructions
Typically, we think of verbs as having a particular argument structure that maps syntactic arguments to semantic roles. This runs into problems with “John played the piano to pieces”. The verb “play” can be used intransitively or transitively, but not typically in this way. Instead, proposed a RESULTATIVE construction that changes the valency of arbitrary verbs, and also applies to “He licked the spoon clean” as well as nonsense Jabberwocky words.
Several more valency-increasing constructions. The DITRANSITIVE construction is eg: “I gave John the keys”, typically the first argument is an agent that transfers an object (second argument) to a recipient (third argument). The agent must have done this willingly, and the recipient must also be willing, or at least “socially qualified”. It can also be used metaphorically and there are certain classes of words that don’t allow this construction like “explain” — although native speakers disagree about which words are allowed.
The CAUSED MOTION construction is eg: “The audience laughed Bob off the stage”. Semantically, the agent causes the theme to move on a path towards a goal. The agent cannot be an instrument, and the path of the theme is usually intended. The WAY construction is eg: “Frank dug his way out of prison”. Semantically, the agent moves along a path that takes some effort to traverse, and must succeed in reaching the goal.
There are valency-decreasing constructions too. The PASSIVE is often analyzed as a grammatical transformation of the active, but there is evidence that these are two different constructions. Not all passive counterparts to active sentences are acceptable, and it’s difficult to characterize when it is so. Also, a few verbs only appear in the passive.
The IMPERATIVE construction omits the agent which is assumed to be “you”. The requests are typically dynamic actions rather than states. Pragmatically, the function is not always to give orders, but a polite discourse marker. Some constructions like LABELESE only apply to very specific communicative situations, like “Made in China”, another is cookbook instructions like “Fry until lightly brown”, this would be odd if spoken.
Syntactic alternations are when two surface forms seem to mean the same thing. However, there are usually semantic differences between the alternations nd it’s not always the case that both are acceptable, thus construction grammar advocates for analyzing them as separate constructions. The semantic coherence principle says that a construction may only be applied if the result makes sense semantically.
Ch3: The construct-i-con
A challenge to construction grammar comes from meaningless constructions, like SUBJECT-PREDICATE or MODIFIER-HEAD. These seem like formal criteria (eg: the subject must agree in person and number with the main verb) without any common semantic core, as different instances of these constructions share little semantically. We must as ourselves what is the structure of the construct-i-con.
Inheritance is the idea that constructions are in a hierarchy with the most abstract like SUBJECT-PREDICATE at the top. Several types of inheritance links: instance link is when “face the music” is a construction that is a special case of TRANSITIVE. Polysemy links is when the meaning of a construction is extended, like “John’s book” is the prototypical genitive construction indicating possession, but in “John’s train”, John does not own the train. Constructions may be similar to each other in form or meaning, like TRANSITIVE and DITRANSITIVE, these are subpart links. In an utterance, multiple constructions are invoked simultaneously and in a network.
Constructional contamination is when other constructions influence a speaker’s choice of words. For example, speakers prefer “the virus was sexually transmitted” over “transmitted sexually” because of the frequent phrase “sexually transmitted diseases”. Linguists disagree about whether inherited information is stored once at the top level (complete inheritance) or replicated at all levels (redundant representations).
Are constructions just another name for phrase structure rules in traditional syntax? Not exactly, because the perspectives are different. In traditional syntax, noun phrase is a fundamental building block, but in constructional grammar, constructions are the fundamental building block, and traditional categories like noun phrase or subject are just generalizations of commonalities across different constructions (like subject in TRANSITIVE and DITRANSITIVE).
Ch4: Constructional Morphology
Construction grammar can be applied to morphology in a similar way as in syntax. The most classic example of a word formation process is wug -> wugs for plural, other common ones include -ed and -able. Not all morphology is concatenative: examples are blending (breakfast + lunch -> brunch) and phrasal compounds (eg: “a don’t ask don’t tell policy”).
Morphemes differ on how productive they are. Some are very productive like -er and -ness, others are unproductive like -ship and -th in warmth. Should these still count as constructions? A lot of morhology have non-compositional meanings like “doctor-recommended” — can mean recommended for doctors or recommended by doctors, thus giving evidence that morphology is a made up of constructions.
Some combinations of morphemes are best analyzed as a single construction like decaffeinate <- caffeine as [de X ate] = to remove X, since “caffeinate” is rare. Affixes can only be ordered in certain combinations, so -ise followed by -ive like “specialisive” is not allowed. The level-ordering hypothesis says there are two levels of affixes, level 1 may serve as input to level 2 but not vice-versa.
Why does this ordering exist? One theory is the complexity-based ordering hypothesis, which says some affixes are closely attached so they’re not easily parsed out, eg: “government” is usually thought of as one word rather than having a -ment affix (this is called entrenchment). Affixes that are less entrenched tend to combine more freely with arbitrary inputs, ad appear as an outer level morpheme.
Noun-noun compounding is sometimes viewed as a fairly regular process: endocentric compounding is when [A B] means a type of B that has some relation to A. Still, there are some irregularities. The non-head can only sometimes be plural, like “car factory” but not “cars factory”, except “arms factory” is ok. This is because the singular for “arms” is unavailable. In attributive compounds, an “electrical engineer” is not an engineer that is electrical, but it is someone who does electrical engineering. The parallel here is a type of subpart link.
Ch5: Information Packaging Constructions
There are often many syntactic ways to convey the same meaning. Information packaging constructions don’t add new meaning, but rearrange existing meaning in some way. For example, the WH-CLEFT construction “What John lost was his wallet” is appropriate when contrasting wallet with something else, otherwise you just say “John lost his wallet”.
According to Lambrecht, information packaging constructions are sentence level, and require the speaker to keep track of the listener’s current mental state. Pragmatic presuppositions are what the listener already knows or can take for granted before the utterance, pragmatic assertions are what he knows or can take for granted afterwards. For example, the TOUGH-RAISING construction (“X is hard to understand”) usually has X being old information. A previous utterance can activate a set of concepts, without explicitly mentioning them, eg: “Mary is pregnant again” activates the concepts of baby and father.
Cleft constructions (IT-CLEFT and WH-CLEFT) are used to shift focus to the focus phrase (“what I saw was X” vs “it was X that I saw”). However, difference is IT-CLEFT has a stronger requirement that the presupposition is activated in the listener.
Wh-island constraints are cases where wh-movement is not allowed out of an “island”, determined by syntactic and semantic properties. One analysis says that the gap must not be situated in a part of a construction that’s typically reserved for pragmatically presupposed information. Awareness of information constraints of constructions is a part of a speaker’s knowledge of a language.
Ch6: Constructions and Language Processing
This chapter talks about psycholinguistic evidence for construction grammar. Bencini and Goldberg (2000) asked participants to cluster sentences into 4 groups, and found they clustered more by construction (eg: transitive vs caused-motion) than by the main verb.
An important question is how do children learn that some constructions are ungrammatical without negative input. Boyd and Goldberg (2011) investigated a class called A-adjectives (eg: afraid, alive) which can only be used predicatively and not attributively, but it is impossible to determine whether an adjective is an A-adjective by semantic or phonological factors. They found evidence for statistical preemption when learning novel words like “adax”: if the experimenter used it predicatively (“the cat that is adax moved”), then the participant recognizes it as a hint that it’s an A-adjective.
Gahl and Garnsey (2004) found that people reduced verbs (said them faster) when used in more common constructions. This provides evidence that speakers have some awareness of co-occurrence probabilities between verbs and constructions.
Ch7: Constructions and Language Acquisition
This chapter surveys a bunch of research that finds evidence for construction grammar by observing how children learn language. The continuity hypothesis says that children acquire more or less the same categories and rules as in adult language. In contrast, item-based learning is the view that children learn entire phrases before acquiring generalizations.
Evidence supporting item-based learning is that children produce relatively few novel utterances that they haven’t heard before, and they learn many pivotal schemas: fixed patterns that have only 1 slot that can vary.
Gradually, children acquire full-fledged adult constructions. Diessel and Tomasello (2005) found that children had an easier time repeating subject relative clause filler-gaps than genitive relative clauses, indicating that frequency of construction is a factor, and not just distance between filler and gap.
Ch8: Language Variation and Change
Constructions exhibit a lot of variation even within a speaker, sometimes two constructions are both acceptable, with gradient levels of acceptability. Labov (1969) found that in AAVE, whether the copula is used is a variable rule that depends probabilistically on the ethnicity of the audience group.
The dative alternation is an extremely well-studied case of language variation, similar to the fruit fly in biology. Bresnan (2007) found that speakers preferred the variation that puts the focus information in sentence-final position. Many other factors affect this alternation, such as the verb, and animacy of the arguments. There are also differences in constructions between different varieties of English.
Constructions also change over time: studies can use apparent-time data (comparing speakers of different ages) or real-time data (comparing text from different periods). Often a construction expands from a specific subcommunity to a larger group. The difference between constructionalization (new construction formation) and construction change is a fuzzy one, because either the form or meaning of the construction may change.
Ch9: Constructions in Spoken Language
Much linguistic work has a written-language bias, whereas spoken language is different in many ways. Spoken language has more right branching and less center embedding, since the latter requires more memory load. Spoken language also has self-repair and co-construction (two speakers collaborating to construct a sentence) — these don’t exist in written language.
Projection is when a speaker begins an utterance that makes the listener expect a followup, eg: when starting with “if…”, listener expects “then…” at some point — a projection-fulfilling element. An expansion is a phrase in between that doesn’t fulfill a projection. A retraction syntactically replaces an earlier element, often to elaborate on a point, and not to self-repair. Hopper (1987) considers these “emergent phenomena” and not necessarily proper constructions.
Apo-koinou constructions are when a single phrase has a double function syntactically, like “this is what I’m talking about is the next project”. The DOUBLE-IS construction is when “is” is repeated: “the problem is is that …”. Hypothesized that this is because the syntax and prosody are misaligned. Collaborative insubordination is when a speaker adds an utterance which is syntactically a subordinate clause to the previous speaker’s utterance.
Ch10: Constructions Across Grammars
With bilingual speakers, similar constructions are often shared between languages. Hoder proposes Diasystematic Construction Grammar, in which diconstructions are constructions that span across languages. For example, English and German NPs have a lot in common, although there are differences which result in negative transfer: speakers producing ungrammatical constructions or avoiding certain acceptable ones. Replica grammaticalization is when bilingual speakers transfer a construction into a parallel one in another language, like Pennsylvania German acquiring the GOING-TO construction from English.
L2 language learners try to transfer similar constructions from their native languages, leading to fossilization of non-native-like patterns even after prolonged periods of input. Herbst (2016) suggested a list of foreign language instruction advice based on construction grammar principles, essentially teach constructions of form-meaning pairs and minimize grammatical terminology.