Deriving the Structure of Variation from the Structure of Non-Variation in the English Dative Alternation∗

sense, depending on the verb and context), and the other of which, the goal (also “recipient” or “indirect object”), denotes the recipient or destination.1 In the double object construction (1), the goal NP immediately follows the verb, and is in turn immediately followed by the theme NP. The prepositional construction (2), by contrast, places the theme NP immediately after the verb, while the goal NP follows as the complement of a prepositional phrase headed by an appropriate preposition. The third construction, the so-called heavy NP shift (3), has the same structure, but with the order of the PP and theme NP reversed. A fourth logically possible construction would be of the form V NPtheme NPgoal, like the double object but with theme and goal reversed, e.g., give it me. This ordering has historical relevance (Curme 1928, Cassidy 1937), and appears in restricted contexts in some non-standard dialects (Murphy 2007); nonetheless it remains quite rare or ill-formed overall, and does not occur in the corpus used here. Its absence will be held as a fact to be derived, rather than assumed a priori. An extensive literature aims to explain the observation that for some combinations of verbs and arguments, constructions (1–3) alternate freely as apparent paraphrases, while with other verbs or arguments such alternation is restricted or outright ungrammatical. In general, the prepositional construction is available for any verb-argument combination. Some verbs, however, resist the double object construction altogether, e.g.: (4) * They returned/donated/revealed [their friends] [the books]. Other verbs permit the heavy NP shift construction with some arguments (typically phonologically “large” or “heavy” themes), but not with others: (5) a. I’m going to reveal [to you] [everything I’ve learned in this business]. b. * I’m going to reveal [to you] [it]. Furthermore, in large corpora of spontaneous speech or writing (cf. Collins 1995, Gries 2003, Bresnan et al. 2007, Bresnan & Nikitina 2007), the usage facts are highly gradient, with verb-argument combinations exhibiting one, two, or all three constructions in systematically different proportions of frequency. It is such usage-frequency data that we seek to address. In the remainder of this paper, section 2 describes the corpus from which the present data derive, and section 3 follows with a brief presentation of Anttila’s (2008) prosodic OT analysis of the alternation, which forms the underlying categorical model from which possible grammars are sampled. Section 4 then shows how the variable usage frequencies are predicted, and quantifies the overall fit of the model. Section 5 offers some concluding remarks. 2. The Blogspot corpus The usage frequency data come from 1,601 sentences of informal written English sampled from free, online weblog and homepage hosting sites (about 80% from blogspot.com and 20% from livejournal.com, geocities.com, and other sources). The corpus is reported on by Anttila (2008) and Anttila et al. (2010). It consists of approximately 100 finite dative constructions found for each of the following sixteen ditransitive verbs: (6) SAMPLED VERBS: assign, award, bring, give, offer, promise, administer, bequeath, concede, convey, deliver, donate, explain, guarantee, recommend, reveal Anttila et al. report that the sampling procedure was to include the first 100 hits containing the relevant ditransitive context from the results of a search for each verb. Several auxiliary frames (will give, have given, etc.) were also included in the search queries, in order to better restrict the results to verbal contexts. Each verb usage sampled in this way was classified as an instance of either the double object, prepositional, or heavy NP shift construction. Table 1 shows the overall corpus frequencies, across all verbs, of the three construction types. It can be seen that at this gross level, the prepositional construction is by far the most common. An agentive subject of the verb is of course also usually present, but will not be relevant here.