Base_JSON/prefixU/json/udw/2021.udw-1.14.json

{
    "paper_id": "2021",
    "header": {
        "generated_with": "S2ORC 1.0.0",
        "date_generated": "2023-01-19T03:11:16.561366Z"
    },
    "title": "Mischievous Nominal Constructions in Universal Dependencies",
    "authors": [
        {
            "first": "Nathan",
            "middle": [],
            "last": "Schneider",
            "suffix": "",
            "affiliation": {},
            "email": "nathan.schneider@georgetown.edu"
        },
        {
            "first": "Amir",
            "middle": [],
            "last": "Zeldes",
            "suffix": "",
            "affiliation": {},
            "email": "amir.zeldes@georgetown.edu"
        }
    ],
    "year": "",
    "venue": null,
    "identifiers": {},
    "abstract": "While the highly multilingual Universal Dependencies (UD) project provides extensive guidelines for clausal structure as well as structure within canonical nominal phrases, a standard treatment is lacking for many \"mischievous\" nominal phenomena that break the mold. As a result, numerous inconsistencies within and across corpora can be found, even in languages with extensive UD treebanking work, such as English. This paper surveys the kinds of mischievous nominal expressions attested in English UD corpora and proposes solutions primarily with English in mind, but which may offer paths to solutions for a variety of UD languages.",
    "pdf_parse": {
        "paper_id": "2021",
        "_pdf_hash": "",
        "abstract": [
            {
                "text": "While the highly multilingual Universal Dependencies (UD) project provides extensive guidelines for clausal structure as well as structure within canonical nominal phrases, a standard treatment is lacking for many \"mischievous\" nominal phenomena that break the mold. As a result, numerous inconsistencies within and across corpora can be found, even in languages with extensive UD treebanking work, such as English. This paper surveys the kinds of mischievous nominal expressions attested in English UD corpora and proposes solutions primarily with English in mind, but which may offer paths to solutions for a variety of UD languages.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Abstract",
                "sec_num": null
            }
        ],
        "body_text": [
            {
                "text": "to space reasons; we hope that guideline discussions in those languages will benefit from the analyses below. 2",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "We turn first to proper names, especially names of persons, and the constructions by which a speaker can elaborate on a nominal referring expression.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Name Descriptors",
                "sec_num": "2"
            },
            {
                "text": "(1) a. I met Gaspard Ulliel.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Name Descriptors",
                "sec_num": "2"
            },
            {
                "text": "b. I met Gaspard Ulliel, the French actor. c. I met the French actor, Gaspard Ulliel.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Name Descriptors",
                "sec_num": "2"
            },
            {
                "text": "(2) a. I met French actor Mr. Gaspard Ulliel. b. *I met French actor. c. *I met the Mr. Gaspard Ulliel.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Name Descriptors",
                "sec_num": "2"
            },
            {
                "text": "How are these handled in UD? The flat relation comes into play for open-class expressions with no clear syntactic head, canonically including personal names like Gaspard Ulliel. A flat structure, by convention, is represented in UD by designating the first word as the head of each of the subsequent words, which attach to it as flat (a \"bouquet\" or \"fountain\" analysis).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Name Descriptors",
                "sec_num": "2"
            },
            {
                "text": "The trouble is that referring expressions may contain descriptors beyond personals. Following the Cambridge Grammar of the English Language (CGEL; Huddleston and Pullum, 2002), we distinguish two types of pre-name descriptors in English: An appellation is a title that would be used to formally address somebody by social status (e.g. occupation or gender), such as Mr. Obama or President Obama. An embellishment 3 is a bare nominal phrase preceding the name (and appellation if there is one) describing the referent with category information like actor, French actor, or surprise winner of the Kentucky Derby. 4 The embellished name may have an inanimate referent, as in German car maker BMW. 5 In English, embellishments are characteristic of select genres such as news. 6 (2a) contains an embellishment and an appellation within the same referring expression. The current UD guidelines state: 7",
                "cite_spans": [
                    {
                        "start": 611,
                        "end": 612,
                        "text": "4",
                        "ref_id": null
                    },
                    {
                        "start": 694,
                        "end": 695,
                        "text": "5",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Name Descriptors",
                "sec_num": "2"
            },
            {
                "text": "If the two nominals participate in denoting one entity, the default relation to connect them is flat (which may also be used to connect other nodes that are not nominals). Typical examples are personal names: we can say that John Smith is a special type of John as well as a special type of Smith, but none of the names governs the other and either of them can be omitted. In many languages this analysis extends to titles and occupations, as in English president Barack Obama.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Name Descriptors",
                "sec_num": "2"
            },
            {
                "text": "Yet the flat analysis for embellishments and appellations yields counterintuitive results. That they are bare NPs and are omissible-whereas the personal name is not, as shown by (2b)-is strong syntactic evidence that they are modifiers. Moreover, it should be intuitively obvious that Gaspard and Ulliel form a coherent unit of structure-yet under the bouquet analysis for flat structures (i.e. attaching all children to the first token), Ulliel would have distinct heads for Gaspard Ulliel, French actor Gaspard Ulliel, and Mr. Gaspard Ulliel. 8 Further discussion in the guidelines acknowledges treating titles as flat is controversial, but explains that titles do not meet typical criteria for nmod, compound, or appos. An nmod typically receives its own independent case marking (possessive or prepositional in English). appos is limited in UD to relations : An appellation with current vs. proposed structures (several options for the relation label of the \"appellation\" dependency discussed below).",
                "cite_spans": [
                    {
                        "start": 545,
                        "end": 546,
                        "text": "8",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Name Descriptors",
                "sec_num": "2"
            },
            {
                "text": "between two full NPs (or DPs, i.e. NPs including a determiner), as in (1b, 1c). And crosslinguistically, \"titles do not usually behave like compounds: in German, they are not joined to the following words, as compounds are normally joined in German, and they appear at the beginning of names in both German and Hebrew, even though German compounds are head last and Hebrew compounds are head first.\" 9 Nevertheless, we suggest that appellations and embellishments be removed from the flat analysis. Exactly how this could be achieved is considered below.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Name Descriptors",
                "sec_num": "2"
            },
            {
                "text": "A narrow solution would be to group appellations and embellishments under the category of titles. As these constructions are frequent and distinctive, a subtype called :title might be appropriate, and subtyping could alleviate the concern that none of the existing top-level deprels is a perfect fit. Alternatively, a new top-level relation could be introduced. We thus begin by considering the following options:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "A Relation for Titles?",
                "sec_num": "2.1"
            },
            {
                "text": "\u2022 title, a new top-level relation \u2022 compound:title \u2022 appos:title \u2022 nmod:title \u2022 nmod:desc, a broader subtype, meant to cover additional mischievous nominals A new top-level relation? A new top-level (universal) relation, title, presupposes that honorific titles, at least, occur widely across languages and may have idiosyncratic syntax. However, it seems possible that in some languages titles might have 'normal' syntax, and would not need such a top-level relation at all. Even for languages with conspicuous title syntax, UD relations aim to be as compact as possible; adding major labels is not done lightly, and would require waiting for UDv3, not to mention imposing costs on many treebank maintainers and requiring updates to existing tools. We therefore prefer subtyping an existing relation. Problems with compound:title. In English, compound dependent nouns too are bare (lack a determiner of their own), similar to appellations and embellishments, suggesting a subtype compound:title. In fact, there is prior art in UD: Finnish UD documents the label compound:nn for appellations. 10 However, there are important differences that suggest compound nominals (at least in English) and titles are two different beasts. While the definition of compound is quite vague, its applicability to modifiers of nouns is clearest in determinative compounds, either where both the head and modifier are part of a multiword proper name like Washington Post; or where the head denotes a kind (usually, a noun that could be made either definite or indefinite) which is restricted by the modifier, e.g. cake flavors. Often such non-name combinations could be paraphrased with a possessive or prepositional construction if used literally (flavors of cake); and often compounds behave like complex words and may become lexicalized as idiomatic multiword expressions. By contrast, appellations and embellishments of proper name heads nonrestrictively add information about an entity and might be paraphrased with \"who is\" or an appositive (French actor Gaspard Ulliel \u2192 Gaspard Ulliel, the French actor / Gaspard Ulliel, who is a French actor).",
                "cite_spans": [
                    {
                        "start": 1093,
                        "end": 1095,
                        "text": "10",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "A Relation for Titles?",
                "sec_num": "2.1"
            },
            {
                "text": "Morphosyntactic evidence also weighs against the compound analysis: English compound modifiers are very rarely plural, even when denoting multiple items-whereas appellations, embellishments, and appositives agree in number with their referent:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "A Relation for Titles?",
                "sec_num": "2.1"
            },
            {
                "text": "(3) a. Presidents Obama and Biden [appellation] 9 https://universaldependencies.org/u/dep/flat.html#some-further-notes-on-relations-for-names Cartons of/for multiple eggs are egg cartons, stripping the plural ending from the compound modifier. 12 If appellations and embellishments were special cases of the English compound construction we would expect them to resist pluralization as well, but this is not the case (3a, 3b).",
                "cite_spans": [
                    {
                        "start": 34,
                        "end": 47,
                        "text": "[appellation]",
                        "ref_id": null
                    },
                    {
                        "start": 244,
                        "end": 246,
                        "text": "12",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "A Relation for Titles?",
                "sec_num": "2.1"
            },
            {
                "text": "Problems with appos:title. Part of the practical motivation for the appos relation is to express a semantic notion of equivalence between referring expressions, such that an information extraction system could strip out supplementary information when matching names against entities in a knowledge base. Thus French actor Gaspard Ulliel, my hero since childhood, won an Oscar could be simplified to Gaspard Ulliel won an Oscar by removing appos and appos:title dependents. From an argument structure perspective, appos is characterized by not adding participants to valency frames, i.e. Gaspard Ulliel and my hero since childhood both instantiate the subject of won.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "A Relation for Titles?",
                "sec_num": "2.1"
            },
            {
                "text": "On the other hand, appos is already rather complicated (spelled out in detail below, \u00a72.3). While embellishments are sometimes categorized as appositions, there is a lack of universal agreement that appellations and embellishments qualify as appositive modifiers; other sources (e.g., Ruppenhofer et al., 2016, p. 77) view the name rather than the embellishment as the appositive phrase.",
                "cite_spans": [
                    {
                        "start": 285,
                        "end": 317,
                        "text": "Ruppenhofer et al., 2016, p. 77)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "A Relation for Titles?",
                "sec_num": "2.1"
            },
            {
                "text": "Intermediate Proposal: a subtype of nmod. The rationale here is that nmod is the most general relation for nominals modifying other nominals. (It already has subtypes, including nmod:poss for possessive modifiers and nmod:tmod for temporal modifiers.) In English, plain nmod dependents have case marking or prepositions, but the subtyping can signal a morphosyntactically exceptional construction, as is already the case with prepositionless nmod:tmod.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "A Relation for Titles?",
                "sec_num": "2.1"
            },
            {
                "text": "If we target only titles, then nmod:title is the least objectionable solution narrowly tailored for embellishments and appellations, given that (a) nmod already has other subtypes, (b) this would avoid confusion with dominant uses of compound and appos, and (c) implementing a new universal relation across treebanks would be onerous, but treebanks are allowed flexibility to diverge and innovate with subtypes. On the other hand, there are a number of other 'mischievous' adnominal constructions requiring a solution, which suggests that a subtype focusing only on titles may be too narrow, motivating a more general name fitting other types of descriptive modifiers, for which we will propose a new relation (called nmod:desc).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "A Relation for Titles?",
                "sec_num": "2.1"
            },
            {
                "text": "The above discussion is limited to appellations and embellishments that precede a name. But other, less frequent constructions bear some resemblance to these:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Other Special Types of Nominal Modification",
                "sec_num": "2.2"
            },
            {
                "text": "(4) Post-name bare nominal modifiers: a. 11-year-old Draco, scion of the Malfoy family, was sorted into Slytherin. b. Oedipus, King of Thebes (5) First or second person pronoun plus noun: 13 a. We pilots deserve a pay raise. b. You guys deserve a pay raise. 14 In (4, 5), the bolded nominal phrase can be omitted while its head (underlined) cannot. (4a) can be considered a post-head embellishment, and (4b) a post-head appellation. The construction seen in (5),",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Other Special Types of Nominal Modification",
                "sec_num": "2.2"
            },
            {
                "text": "headed by a pronoun, is a cousin of the pre-head embellishment, as shown by the third person paraphrase of (5a): pilots Earhart and Lindbergh. A broad relation nmod:desc for the special cases seen above as well as appellations and embellishments would separate them from the appos, compound, and flat cases while covering sufficient ground to merit its inclusion.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Other Special Types of Nominal Modification",
                "sec_num": "2.2"
            },
            {
                "text": "A classic example of an appositive appears in (6). The appositive phrase, my brother, is a nonrestrictive full NP descriptor of Sam. It is syntactically omissible, and could in fact replace its head as they share the same referent. A similar phenomenon appears in 7, where an indefinite NP ascribes a property to Sam:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "More on Appositives",
                "sec_num": "2.3"
            },
            {
                "text": "(6) Sam, my brother, is very tall.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "More on Appositives",
                "sec_num": "2.3"
            },
            {
                "text": "(7) Sam, a musician, is very tall.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "More on Appositives",
                "sec_num": "2.3"
            },
            {
                "text": "The current definition of the appos relation establishes the following criteria: While appositive phrases are often separated by commas or parentheses, this is not a strict requirement, and of course spoken language has no commas. We understand the definition to also include:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "More on Appositives",
                "sec_num": "2.3"
            },
            {
                "text": "(9) a. my brother Sam b. the color purple c. the word \"terrorist\" d. the play Much Ado About Nothing",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "More on Appositives",
                "sec_num": "2.3"
            },
            {
                "text": "Cases resembling appositives in some but not all of the above respects require clarification. The bare modifiers discussed above are sometimes considered appositives, but UD excludes them with criterion (8a). (10) satisfies criteria (8a, 8c) but not (8b, 8d), whereas (11) satisfies (8a, 8b, 8d) but fails (8c):",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "More on Appositives",
                "sec_num": "2.3"
            },
            {
                "text": "(10) \"Maybe she really does just need a little space. . . ,\" Amy said, ever the optimist. 16 (11) A new Pakistani leader, he is intent on instituting reforms.",
                "cite_spans": [
                    {
                        "start": 90,
                        "end": 92,
                        "text": "16",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "More on Appositives",
                "sec_num": "2.3"
            },
            {
                "text": "There seem to be two ways forward:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "More on Appositives",
                "sec_num": "2.3"
            },
            {
                "text": "\u2022 Relax appos criteria either in general or in a subtype. In particular, relaxing (8b-8d) would allow appos to cover (10, 11) . This would contrast with nmod:desc suggested above, which covers bare nominal modifiers. \u2022 Maintain the appos criteria in (8), and classify examples such as (10, 11) as dislocated. These constructions are not quite classic dislocation constructions, 17 but they could be treated as if removed from their normal apposition location.",
                "cite_spans": [
                    {
                        "start": 117,
                        "end": 121,
                        "text": "(10,",
                        "ref_id": null
                    },
                    {
                        "start": 122,
                        "end": 125,
                        "text": "11)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "More on Appositives",
                "sec_num": "2.3"
            },
            {
                "text": "In the interest of maintaining the status quo for appositions, we favor the latter solution and recommend using dislocated. 3 Further Issues with Names",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "More on Appositives",
                "sec_num": "2.3"
            },
            {
                "text": "Several other aspects of the syntax of names need to be addressed. The syntactic properties of many of the constructions at issue are summarized in table 1. We begin by underscoring UD's policy of analyzing the internal structure of names with ordinary syntax where possible, regardless of the semantic status of the name. For example, Church Street is analyzed with compound; and New York City consists of an adjective which modifies a noun (amod), which in turn modifies another noun (compound). 18",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Syntactically analyzable proper names",
                "sec_num": "3.1"
            },
            {
                "text": "Cardinal direction modifiers of nouns (north, northeast, etc.) are annotated inconsistently in English UD corpora. Based on the tagging tradition of LDC corpora, these should be treated as nouns unless they bear overt adjectival morphology (northern, etc.). Cardinal direction nouns premodifying nouns should therefore attach as compound, whether the expression is a proper name (North Carolina) or not (north coast). When multiple parts of a cardinal direction term are separated by a space or hyphen, they are joined with compound: e.g. north east 'northeast'.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Cardinal directions",
                "sec_num": "3.2"
            },
            {
                "text": "Many proper names incorporate a transparent entity type. In the Thames River, the name is constructed as an ordinary endocentric compound, with the entity type last and serving as the head and an identifier as the modifier. 19 But the River Thames (along with the other examples in (12)) poses a problem as the order is reversed: and Pullum, 2002, pp. 519-20) . However, this omission of the entity type could be viewed as a shortening not unlike reducing Fenway Park to Fenway on the assumption that the speaker is able to identify the referent based on the more specific part of the name. Such shortenings will vary in felicitousness depending on the particular name and context. (Plain Michigan does not refer to the same thing as Lake Michigan.) Note also that the name-initial entity types may be pluralized when grouping together multiple entities of the same type, which distinguishes them from flat structures or typical compound modifiers and suggests they may be heads: Lakes Michigan and Ontario (cf. Mirror and Swan Lakes). This fits with the expected semantics, as noun-noun compounds tend to be headed by the superordinate category, and historically it is possible that the construction is in fact a remnant of left-headed compounding from Romance place names, possibly from Norman toponym patterns (English Mount X, French Mont-X, e.g. Mont-Saint-Michel).",
                "cite_spans": [
                    {
                        "start": 224,
                        "end": 226,
                        "text": "19",
                        "ref_id": null
                    },
                    {
                        "start": 330,
                        "end": 359,
                        "text": "and Pullum, 2002, pp. 519-20)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Names beginning with an entity type",
                "sec_num": "3.3"
            },
            {
                "text": "We therefore consider the examples in (12) as inverted (left-headed) compounds. 20 The identifier can attach to the entity type as compound to reflect the inverted word order in these kinds of names.",
                "cite_spans": [
                    {
                        "start": 80,
                        "end": 82,
                        "text": "20",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Names beginning with an entity type",
                "sec_num": "3.3"
            },
            {
                "text": "Numbers can also figure into names. They can disambiguate multiple of a series of related entities named by a proper noun, as in (13). These are appendages to a proper name, syntactically omissible (with a resulting broadening of meaning), and could be treated as modifiers. Numbers can also follow an entity type, as in (14). The cases in (14) use the number to identify a specific instance of the type. The entity type appears first, similar to the inverted compound examples in \u00a73.3. It is a completely different construction from quantity modification, the predominant application of nummod, as in 3 items (plural!) or 3%. A morphosyntactic difference between the numeric modifier constructions in (13) and (14) is that only the latter exhibit agreement: page 5 (one page), pages 5-10 (multiple pages), but *Firefoxes 58.0 and 59.0.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Numbered entities",
                "sec_num": "3.4"
            },
            {
                "text": "We see three options, each with pros and cons: \u2022 The morphosyntactic difference notwithstanding, treat (13) and (14) as essentially the same construction, with a new relation such as nummod:name (consistent with the fact that the superordinate category nummod is currently applied to numeric modifiers generally). 21 Advantages are that (13) and (14) look very similar, and numbers are a salient property for annotators or corpus users to notice when selecting the appropriate relation. However, adding a subtype for a relatively narrow and infrequent phenomenon is questionable, and some cases are not numeric (Level B). \u2022 Treat (13) and 14as instances of more general constructions. The construction in (14) can be considered an inverted compound like Lake Michigan ( \u00a73.3). Flat structures could apply to the names in (13) as this construction is less morphologically transparent. This would avoid a new subtype but also may be seen as splitting hairs based on a subtle morphosyntactic criterion. \u2022 A third option is to adopt nmod:desc for the constructions in (13) and (14). This would essentially restrict the definition of compound to substantive lexical material excluding numbering designators; nmod:desc would broadly cover miscellaneous modifiers associated with names that do not fit the more conventional constructions. This solution eclipses the similarity between Lake Michigan (which would remain compound) and Figure 4 , but it perhaps avoids a counterintuitively broad application of compound. It also means that the scope of nmod:desc is a bit broader, including not just modifiers that are secondary to the main part of a name, but also modifiers that are essential to it (just Figure is not a name, whereas Ulliel is). (13a, 14e, 14f) illustrate a construction in which a word like number or version may precede a number to clarify that it is an identifier rather than a quantity. In modern usage this would generally remain singular even if referring to multiple items (items number 3 and 4), so we analyze number as a compound modifier by default, and nmod:desc only if plural (items numbers 3 and 4). \"?\" is provided as a stand-in for the relation between the entity type and the number given the above uncertainty: 22",
                "cite_spans": [
                    {
                        "start": 314,
                        "end": 316,
                        "text": "21",
                        "ref_id": null
                    }
                ],
                "ref_spans": [
                    {
                        "start": 1426,
                        "end": 1434,
                        "text": "Figure 4",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Numbered entities",
                "sec_num": "3.4"
            },
            {
                "text": "No. For hyphenated numeric ranges (14c), the prevailing policy in UD corpora has been to analyze the second part like a prepositional phrase to 10, thus an nmod of 5. One of the authors takes the view that a coordination analysis would be more natural. In any event, 5 attaches to pp. as a modifier.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Symphony",
                "sec_num": null
            },
            {
                "text": "Adjective-expanding suffixes like Inc. (\"incorporated\") in Apple Inc. should attach as amod. Nominal suffix designations that do not head the name, e.g. LLC (\"limited liability corporation\"), should attach as nmod:desc. For personal names, the suffix type III in (13b) is addressed above. Generational name suffixes that do not use numerals, like Richard Jr. and Richard the Third, are treated as postmodifying amod. Other abbreviated name suffixes that would expand to nominal expressions, such as professional or honorary designations (MD, O.B.E.), attach as nmod:desc.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Business and personal name suffixes",
                "sec_num": "3.5"
            },
            {
                "text": "A nickname that takes the form of a full NP appended to a name, e.g. Richard the Lionheart, can be attached as appos. The same goes for works of art featuring a formulaic name followed by a nickname: Symphony No. 5 \"Fate\". Parenthetical descriptions following a name that are not alternate references to the entity should be treated as parataxis: Pierre Vinken, 61, said. . . ; Vinken, 61 years old, said. . . ; The Chicago Manual of Style, 17th edition; Biden (D) said. . . (but Biden, a Democrat, said. . . would be appos).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Nicknames and parenthetical descriptors",
                "sec_num": "3.6"
            },
            {
                "text": "A street address like 221b Baker St. is headed by St., with Baker attaching as compound, and 221b per the policy on numbered entities ( \u00a73.4). Frequently, place descriptions specify a locale-NP postmodifier without a connective word besides punctuation. Examples: London, UK; University of Wisconsin-Madison; CSI: Miami. These should be considered adverbial NPs, which arguably should fall under the nmod:npmod relation. 23 Multiple tokens of a single phone number should be joined with flat (this is the practice in the GUM corpus; EWT currently favors nummod). Separate pieces of metadata that are juxtaposed in an extralinguistic fashion (e.g., name, street address, city, postal code) should be treated as items of a list-successive items should attach to the first as list.",
                "cite_spans": [
                    {
                        "start": 421,
                        "end": 423,
                        "text": "23",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Addresses",
                "sec_num": "3.7"
            },
            {
                "text": "In English, the attributive modifier position before the noun head in a noun phrase is not limited to adjectives/adjective phrases (very easy to use) and nominals. It also accommodates phrases like: 22 Confirming native speaker intuitions, a search of COCA (Davies, 2010) reveals that the plural is much less frequent than the singular in the pattern N.PL number(s) NUM and NUM, with the exception of the abbreviated spelling, where nos. is more prevalent in this context than no. (the abbreviations seem to be especially conventional in proper names like Symphony No. 5).",
                "cite_spans": [
                    {
                        "start": 199,
                        "end": 201,
                        "text": "22",
                        "ref_id": null
                    },
                    {
                        "start": 257,
                        "end": 271,
                        "text": "(Davies, 2010)",
                        "ref_id": "BIBREF1"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Phrasal Attributive Modifiers",
                "sec_num": "4"
            },
            {
                "text": "23 Currently, corpora sometimes use nmod:npmod and sometimes use appos, which is not appropriate as the two parts of the location are not interchangeable. Space does not permit full discussion of nmod:npmod here (but see Schneider and Zeldes, 2021, \u00a76) . (15) a. a high-quality product b. a by-the-book strategy c. a fly-by-night operation d. a have-your-cake-and-eat-it-too plan e. a come-to-Jesus, do-or-die moment f. a stern don't-mess-with-me look g. a must-see movie h. fire-breathing dragons i. the Bible-thumping, church-going faithful j. many so-called libertarians k. a cost-effective, nuclear-free future",
                "cite_spans": [
                    {
                        "start": 221,
                        "end": 252,
                        "text": "Schneider and Zeldes, 2021, \u00a76)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Phrasal Attributive Modifiers",
                "sec_num": "4"
            },
            {
                "text": "Assuming that the hyphenated expressions are tokenized as separate words, UD annotators are confronted with two issues: how to analyze these phrases internally, and which dependency relation to use for the modification of the external noun.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Phrasal Attributive Modifiers",
                "sec_num": "4"
            },
            {
                "text": "Some of the hyphenated expressions in (15) are clearly lexicalized; others are productive combinations. Expressions of this type might loosely be described as 'compounds', in the sense that the joining of multiple content words into one lexical item is the morphological process of compounding. Should the hyphenated parts thus be joined together with compound across the board? We are hesitant to establish this policy because it would overload an already very broad relation label. Centrally, in noun phrases, compound describes modification of a noun by another noun. If it applies to the examples in (15), it would be for attachment to the underlined noun, not the internal structure of the hyphenated expression.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Phrasal Attributive Modifiers",
                "sec_num": "4"
            },
            {
                "text": "Another consideration is that the internal structure of the hyphenated phrases is largely regular: phrasal modifiers of nouns can be structured as modified nouns (15a), PPs (15b), VPs (15c, 15d), imperative sentences (15e, 15f), and verb clusters (15g). These structures are transparent, and just as UD policy analyzes regular internal structures in proper names like University of Wisconsin, we advocate recognizing internal structure here.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Phrasal Attributive Modifiers",
                "sec_num": "4"
            },
            {
                "text": "Yet synthetic or argument structure compounds such as fire-breathing, Bible-thumping, and churchgoing (15h, 15i) invert the normal clausal order. Neither fire nor Bible nor church is the subject in the clausal paraphrase: fire is the direct object in breathing fire; the paraphrase of Bible-thumping would require reordering and adding a determiner or plural for the direct object; and the paraphrase of churchgoing would require a preposition: going to church. Meanwhile, so-called (15j) lacks any obvious paraphrase as a clause. We take these anomalies in word order and morphosyntax as clear evidence that left-headed 'deep structure' VP material is being grafted onto a right-headed compound in the 'surface structure'. As Basic UD aims to represent surface syntax, we join these expressions as compound, as shown for fire-breathing in figure 2c (vs. figure 2b) . The adjective-headed combinations in (15k) should also use internal compound, as should numeric modifier compounds like a 10-year plan. 24 The next question is the external attachment, which is made difficult by UD's lexicalist principle that the part of speech of a word determines which relations it can participate in. Consider must-see (15g), which is not a full VP, merely an auxiliary plus its head verb. Is this to be treated as a clausal dependent-acl, or even acl:relcl (a relative clause)? This seems dubious; note that a relative clause paraphrase would involve an embedded subject, e.g. a movie that one must see, or else a passive-a movie that must be seen. It is also doubtful whether (15c-15f) should be treated as clausal modification, yielding several different dependency labels for the attributive relationship. A simpler solution, it seems to us, is to treat attributive phrasal expressions internally headed by verbs like coerced noun phrases, 25 with compound for the external attachment, as shown in figure 2a. As for PP modifiers like in (15b), it seems simplest to attach them as compound rather than nmod; on this view, English nominal compound is equivalent to attributive modification by a non-possessive nominal phrase (a hypothetical alternate name being nmod:attr).",
                "cite_spans": [
                    {
                        "start": 1004,
                        "end": 1006,
                        "text": "24",
                        "ref_id": null
                    }
                ],
                "ref_spans": [
                    {
                        "start": 840,
                        "end": 865,
                        "text": "figure 2c (vs. figure 2b)",
                        "ref_id": "FIGREF3"
                    }
                ],
                "eq_spans": [],
                "section": "Phrasal Attributive Modifiers",
                "sec_num": "4"
            },
            {
                "text": "To summarize, our proposed policy for phrasal attributive modifiers of nouns is:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Phrasal Attributive Modifiers",
                "sec_num": "4"
            },
            {
                "text": "\u2022 The attributive expression is internally analyzed with regular relations to the extent possible, except where those relations defy ordinary word order or morphosyntax. compound is used internally for anomalous relations. \u2022 In the interest of simplicity, all non-possessive attributive modifiers attach as either compound if internally headed by a nominal or nominalized phrase (including PPs), and amod etc. for adjectival heads, as appropriate.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Phrasal Attributive Modifiers",
                "sec_num": "4"
            },
            {
                "text": "While analytically expressed dates like the thirty-first of July follow normal syntax (with thirty-first elliptical for thirty-first day), there are special written formats for dates and times. Instead of a flat structure, which would obscure the compositionality of dates, we propose the simple principles of (a) treating the most precise part of the expression as its head, and (b) connecting the parts of the expression together with nmod:tmod. 26 For example, July 31, 1980 AD consists of a year expression (1980 AD) and a month both modifying a date: Another convention puts the date before the month (31 July). There, too, the date would be the head.",
                "cite_spans": [
                    {
                        "start": 448,
                        "end": 450,
                        "text": "26",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Dates",
                "sec_num": "5"
            },
            {
                "text": "Even when the date is written as an ordinal-July the fourth-the month should be considered a temporal modifier because it can be omitted with sufficient context (I'll see you on the fourth; *I'll see you on July). This is in contrast to Richard the Third ( \u00a73.5), where Richard is the head.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Dates",
                "sec_num": "5"
            },
            {
                "text": "A further practical consideration is that UD tree heads are often used to determine minimal token spans for annotations such as entity recognition, mentions in coreference resolution, and entity linking spans for Wikification (associating mentioned entities with their Wikipedia entries; Ratinov et al., 2011) . Such minimal or 'MIN' spans (Poesio et al., 2018, p. 12) are then used for training and scoring systems in 'fuzzy' match scenarios. It makes intuitive sense for the day in date expressions to form the minimal span which needs to be identified, since the other tokens, i.e. years and months, already form the minimal spans for the nested mentions of those years and months as separate entities. This use of UD-tree heads is already in place for non-UD corpora using UD parses, such as ARRAU (Uryupina et al., 2020) , and in the gold standard UD English GUM for NER, coreference and Wikification (Lin and Zeldes, 2021) .",
                "cite_spans": [
                    {
                        "start": 288,
                        "end": 309,
                        "text": "Ratinov et al., 2011)",
                        "ref_id": "BIBREF12"
                    },
                    {
                        "start": 340,
                        "end": 368,
                        "text": "(Poesio et al., 2018, p. 12)",
                        "ref_id": null
                    },
                    {
                        "start": 802,
                        "end": 825,
                        "text": "(Uryupina et al., 2020)",
                        "ref_id": "BIBREF16"
                    },
                    {
                        "start": 906,
                        "end": 928,
                        "text": "(Lin and Zeldes, 2021)",
                        "ref_id": "BIBREF5"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Dates",
                "sec_num": "5"
            },
            {
                "text": "For time expressions we follow similar reasoning, with an example as follows:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Dates",
                "sec_num": "5"
            },
            {
                "text": "10:00 pm UTC nmod:tmod",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Dates",
                "sec_num": "5"
            },
            {
                "text": "The time zone could alternately be expressed as a phrase like London time, which we would also view as nmod:tmod. If written as ten o'clock, the token o'clock is considered an adverb and advmod of ten. This also corresponds to an etymological reading of o'clock (< of clock), since a univerbized prepositional phrase is equivalent to an adverb (cf. adverbs like ashore, formed with the Old English preposition an, the stressed equivalent of on). Zeman (2021) likewise proposes a standard for dates and times (considering English as well as Czech, Indonesian, and Chinese). That approach is similar, differing mainly in treating the year in a date expression as headed by the month rather than the date-1980 would be a dependent of July, which would be a dependent of 31, in July 31, 1980. While semantically intuitive (smaller units of time head the next larger containing unit), it is not clear that there is any syntactic motivation to group the month and year together. Although the month cannot normally be omitted while retaining the year, an expression like the 31st, 1980 is only semantically nonsensical, or at best pragmatically anomalous, but not truly ungrammatical. As evidence for this we consider the possibility of felicitous day+year expressions, such as New Year's Day 2000 (the same as 2000-01-01) or Pentecost 2022 (2022-06-05). The year-modifies-month approach also has the disadvantage of creating nonprojectivity if the date is written between the month and the year.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "nmod:tmod",
                "sec_num": null
            },
            {
                "text": "Zeman ( \u00a75) suggests appos to link a date with a day of the week, as in Wednesday, July 31. We agree with this policy. Though the day of the week conventionally comes first in English, we recognize that the order may be reversed on occasion (reversibility is a definitional criterion for appos, which is always left-headed). Moreover, this does not affect preposition choice, as on marks days of the week as well as dates, supporting the appos analysis in which they are essentially interchangeable full NPs.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "nmod:tmod",
                "sec_num": null
            },
            {
                "text": "6 How prevalent are these issues?",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "nmod:tmod",
                "sec_num": null
            },
            {
                "text": "Some readers may wonder how common the issues raised in this paper actually are, and in particular whether their frequency merits adding relation subtypes such as nmod:desc. Table 2 gives statistics for some types of constructions that would be covered under the umbrella of such a relation. Although the phenomena are not extremely frequent, the total token count of 373 out of 152K tokens in the UD v2.9 edition of GUM puts a putative relation covering these at rank 35 of 49 relation labels (including subtypes), between obl:tmod (362 tokens) and nmod:tmod (399), suggesting that these are not particularly rare occurrences. We also presume that depending on genre, some subtypes may become much more frequent, such as company suffixes or even personal titles-for example, the frequency of just company suffixes in EWT seems is about 2.5 per 10K tokens, compared to 0.3 per 10K tokens in GUM (other categories are harder to identify, since their annotation in EWT currently varies or is not easily distinguishable, as in the case of numbering modifiers). Although adding a new labeling distinction in the form of nmod:desc would doubtless require some manual disambiguation effort, we feel that by surveying the constructions in this paper in detail, it becomes more feasible to design high recall, automatic approaches to creating an initial updated version of UD English with a more nuanced treatment of these mischievous constructions, using UD editing libraries such as DepEdit (Peng and Zeldes, 2018) or Udapi (Popel et al., 2017) , which can then be subjected to a manual filtering pass.",
                "cite_spans": [
                    {
                        "start": 1485,
                        "end": 1508,
                        "text": "(Peng and Zeldes, 2018)",
                        "ref_id": "BIBREF9"
                    },
                    {
                        "start": 1518,
                        "end": 1538,
                        "text": "(Popel et al., 2017)",
                        "ref_id": "BIBREF11"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 174,
                        "end": 181,
                        "text": "Table 2",
                        "ref_id": "TABREF6"
                    }
                ],
                "eq_spans": [],
                "section": "nmod:tmod",
                "sec_num": null
            },
            {
                "text": "Above we have reviewed many constructions involving names, values, and compounds that have pointed to blind spots in the current guidelines for the nmod: * , compound, flat, appos, and nummod relations. We have laid out several options for improving the treatment of these constructions via clearer and more principled guidelines. The proposed improvements are of a surgical nature, minimizing disruption to other UD conventions (no new universal relations are proposed, for instance). We are cognizant that considerable effort may be required to fully revise existing UD treebanks, but note that treebanks are already inconsistent; clearer guidance can only help. Subtypes remain officially optional-it is not necessary for a treebank to distinguish subtypes of nmod to be compliant with the UD standard.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusion",
                "sec_num": "7"
            },
            {
                "text": "We invite feedback on these proposals from the UD community, particularly with regard to other languages. We are aware that treebanking efforts in other languages have encountered some of the same issues, but we have not systematically investigated our proposed solutions beyond English.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusion",
                "sec_num": "7"
            },
            {
                "text": "An extended version of this paper(Schneider and Zeldes, 2021) contains additional recommendations regarding numbers and adverbial NPs, omitted here due to space limitations.3 Also called \"false title\", described here as a kind of apposition: https://en.wikipedia.org/wiki/False _ title 4 An anonymous reviewer has commented on the difficulty of applying the bare nominal diagnostic in languages with different determiner systems, such as Slavic languages, Chinese, or Japanese. We fully acknowledge that equivalent constructions may look quite different in those languages, but also believe that the problems analyzed here are both substantial enough in English to merit a more detailed treatment, and common enough in other languages that the discussion is likely relevant beyond English.5 Thanks to an anonymous reviewer for this example. 6 A newscaster might say, Surprise winner of the Kentucky Derby American Pharaoh received a hero's welcome upon returning home today. . . . Note the lack of an article at the beginning of the sentence.7 https://universaldependencies.org/workgroups/newdoc/two _ nominals.html 8 Note that some embellishments and appellations contain clear internal structure (e.g., French actor Ulliel-amod; Secretary of State Clinton-nmod, case). This does not pose an additional problem for the flat analysis, however: even dependents within a flat structure may host internal modifiers, as was recently clarified in the guidelines.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "https://universaldependencies.org/docs/fi/overview/specific-syntax.html#appositions-and-appellation-modifiers",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "It is unclear whether non-coordinated names referring to multiple individuals could license plural embellishments via semantic number agreement: An argument broke out between married actors Brad and Angelina / ?married actors Brangelina / ?British comedians Monty Python.12 For exceptional pluralized modifiers in Germanic compounds see alsoFuhrhop (1996).13 Elsewhere the pronouns are analyzed as determinatives(Huddleston and Pullum, 2002, p. 374), but we deem it impractical to extend det to include such specialized uses of personal pronouns.14 The expression you guys has been conventionalized in some dialects as a gender-neutral second person plural.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "An exception to this constraint is already found in languages with so-called Wackernagel particles, such as Classical Greek or Coptic, which appear in the second position in the sentence and can interrupt any phrase or dependency; seeZeldes and Abrams (2018).16 The Body in the Casket: A Faith Fairchild Mystery, Katherine Hall Page, 201717 The preferatory appositive in (11)-which features a description followed by a definite NP, and would be perfectly at home in a newspaper-is not to be confused with hanging topic left-dislocation with a pronoun referring back to the dislocated element, as might be uttered in conversation: My dad, he is always running late.18 Previously, POS tags in the English treebanks followed Penn Treebank tags and treated all content words within a proper name as PROPN, but this was changed in v2.8; PROPN is now limited to nouns.19 Other place names headed by an entity type and exhibiting ordinary syntax include Mirror Lake, Ford's Theatre, and the Dome of the Rock.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "Another analysis we considered was to treat the entity type as an nmod:desc modifier, giving Lake Michigan the same structure as Dr. Livingstone or actor Ulliel. But the entity types in (12) seem more central to the name than titles, and are not as freely omissible, so we are not persuaded that they are modifiers.21 The choice of subtype parallels flat:name-an optional subtype not currently implemented in English corpora, though it is used for a number of corpora in other languages. The flat:name guidelines currently include Formula 1 as an example; this would become nummod:name in this option.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "Contrast 10-year (compound) with 10 years (nummod), where the number modifier controls agreement.25 Kahane et al. (2017) suggest expanding the UD notion of multiword token to include idiomatic phrasal expressions, separating their external syntactic behavior from their internal structure. This would make it convenient to represent the expression must-see as a multiword NOUN comprised internally of an AUX and a VERB. This could be indicated via a morphological feature ExtPos=NOUN on the internal head, see.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "We considered finer-grained relations like nmod:month, nmod:year, nmod:era, nmod:ampm, and nmod:tz but concluded these were too detailed for UD and should fall under the purview of information extraction.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            }
        ],
        "back_matter": [
            {
                "text": "Though the synthesis of problems, argumentation, and recommendations presented above are new, many of the specific challenges and alternatives arose from background discussions on GitHub about the UD guidelines. These discussions took place over several years in more than a dozen discussion threads involving (at least): Aryaman Arora, Colin Batchelor, Flavio Massimiliano Cecchini, Xinying Chen, \u00c7agr\u0131 \u00c7\u00f6ltekin, Sylvain Kahane, Chris Manning, Bohdan Moskalevskyi, Joakim Nivre, Martin Popel, Alexandre Rademaker, Livy Real, Jack Rueter, Sebastian Schuster, Francis Tyers, Jonathan North Washington, and Dan Zeman. We are grateful to all who provided insights.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Acknowledgments",
                "sec_num": null
            }
        ],
        "bib_entries": {
            "BIBREF0": {
                "ref_id": "b0",
                "title": "A cross-genre ensemble approach to robust Reddit part of speech tagging",
                "authors": [
                    {
                        "first": "Shabnam",
                        "middle": [],
                        "last": "Behzad",
                        "suffix": ""
                    },
                    {
                        "first": "Amir",
                        "middle": [],
                        "last": "Zeldes",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Proc. of the 12th Web as Corpus Workshop",
                "volume": "",
                "issue": "",
                "pages": "50--56",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Shabnam Behzad and Amir Zeldes. 2020. A cross-genre ensemble approach to robust Reddit part of speech tagging. In Proc. of the 12th Web as Corpus Workshop, pages 50-56, Marseille, France.",
                "links": null
            },
            "BIBREF1": {
                "ref_id": "b1",
                "title": "The Corpus of Contemporary American English as the first reliable monitor corpus of English",
                "authors": [
                    {
                        "first": "Mark",
                        "middle": [],
                        "last": "Davies",
                        "suffix": ""
                    }
                ],
                "year": 2010,
                "venue": "Literary and Linguistic Computing",
                "volume": "25",
                "issue": "4",
                "pages": "447--464",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Mark Davies. 2010. The Corpus of Contemporary American English as the first reliable monitor corpus of English. Literary and Linguistic Computing, 25(4):447-464.",
                "links": null
            },
            "BIBREF2": {
                "ref_id": "b2",
                "title": "Deutsch -typologisch",
                "authors": [
                    {
                        "first": "Nanna",
                        "middle": [],
                        "last": "Fuhrhop",
                        "suffix": ""
                    }
                ],
                "year": 1996,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "525--550",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Nanna Fuhrhop. 1996. Fugenelemente. In Ewald Lang and Gisela Zifonun, editors, Deutsch -typologisch, pages 525-550. de Gruyter, Berlin.",
                "links": null
            },
            "BIBREF3": {
                "ref_id": "b3",
                "title": "The Cambridge Grammar of the English Language",
                "authors": [],
                "year": 2002,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Rodney Huddleston and Geoffrey K. Pullum, editors. 2002. The Cambridge Grammar of the English Language. Cambridge University Press, Cambridge, UK.",
                "links": null
            },
            "BIBREF4": {
                "ref_id": "b4",
                "title": "Multi-word annotation in syntactic treebanks -Propositions for Universal Dependencies",
                "authors": [
                    {
                        "first": "Marine",
                        "middle": [],
                        "last": "Sylvain Kahane",
                        "suffix": ""
                    },
                    {
                        "first": "Kim",
                        "middle": [],
                        "last": "Courtin",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Gerdes",
                        "suffix": ""
                    }
                ],
                "year": 2017,
                "venue": "Proc. of the 16th International Workshop on Treebanks and Linguistic Theories",
                "volume": "",
                "issue": "",
                "pages": "181--189",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Sylvain Kahane, Marine Courtin, and Kim Gerdes. 2017. Multi-word annotation in syntactic treebanks - Propositions for Universal Dependencies. In Proc. of the 16th International Workshop on Treebanks and Linguistic Theories, pages 181-189, Prague, Czech Republic.",
                "links": null
            },
            "BIBREF5": {
                "ref_id": "b5",
                "title": "WikiGUM: Exhaustive entity linking for Wikification in 12 genres",
                "authors": [
                    {
                        "first": "Jessica",
                        "middle": [],
                        "last": "Lin",
                        "suffix": ""
                    },
                    {
                        "first": "Amir",
                        "middle": [],
                        "last": "Zeldes",
                        "suffix": ""
                    }
                ],
                "year": 2021,
                "venue": "Proc. of The Joint 15th Linguistic Annotation Workshop (LAW) and 3rd Designing Meaning Representations (DMR) Workshop",
                "volume": "",
                "issue": "",
                "pages": "170--175",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Jessica Lin and Amir Zeldes. 2021. WikiGUM: Exhaustive entity linking for Wikification in 12 gen- res. In Proc. of The Joint 15th Linguistic Annotation Workshop (LAW) and 3rd Designing Meaning Representations (DMR) Workshop, pages 170-175, Punta Cana, Dominican Republic.",
                "links": null
            },
            "BIBREF7": {
                "ref_id": "b7",
                "title": "Universal Dependencies v2: An evergrowing multilingual treebank collection",
                "authors": [
                    {
                        "first": "Joakim",
                        "middle": [],
                        "last": "Nivre",
                        "suffix": ""
                    },
                    {
                        "first": "Marie-Catherine",
                        "middle": [],
                        "last": "De Marneffe",
                        "suffix": ""
                    },
                    {
                        "first": "Filip",
                        "middle": [],
                        "last": "Ginter",
                        "suffix": ""
                    },
                    {
                        "first": "Jan",
                        "middle": [],
                        "last": "Haji\u010d",
                        "suffix": ""
                    },
                    {
                        "first": "Christopher",
                        "middle": [
                            "D"
                        ],
                        "last": "Manning",
                        "suffix": ""
                    },
                    {
                        "first": "Sampo",
                        "middle": [],
                        "last": "Pyysalo",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Proc. of LREC",
                "volume": "",
                "issue": "",
                "pages": "4027--4036",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Joakim Nivre, Marie-Catherine de Marneffe, Filip Ginter, Jan Haji\u010d, Christopher D. Manning, Sampo Pyysalo, Sebastian Schuster, Francis Tyers, and Daniel Zeman. 2020. Universal Dependencies v2: An evergrowing multilingual treebank collection. In Proc. of LREC, pages 4027-4036, Marseille, France.",
                "links": null
            },
            "BIBREF8": {
                "ref_id": "b8",
                "title": "Universal Dependencies v1: a multilingual treebank collection",
                "authors": [
                    {
                        "first": "Joakim",
                        "middle": [],
                        "last": "Nivre",
                        "suffix": ""
                    },
                    {
                        "first": "Marie-Catherine",
                        "middle": [],
                        "last": "De Marneffe",
                        "suffix": ""
                    },
                    {
                        "first": "Filip",
                        "middle": [],
                        "last": "Ginter",
                        "suffix": ""
                    },
                    {
                        "first": "Yoav",
                        "middle": [],
                        "last": "Goldberg",
                        "suffix": ""
                    },
                    {
                        "first": "Jan",
                        "middle": [],
                        "last": "Haji\u010d",
                        "suffix": ""
                    },
                    {
                        "first": "Christopher",
                        "middle": [
                            "D"
                        ],
                        "last": "Manning",
                        "suffix": ""
                    },
                    {
                        "first": "Ryan",
                        "middle": [],
                        "last": "Mcdonald",
                        "suffix": ""
                    },
                    {
                        "first": "Slav",
                        "middle": [],
                        "last": "Petrov",
                        "suffix": ""
                    },
                    {
                        "first": "Sampo",
                        "middle": [],
                        "last": "Pyysalo",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "Proc. of LREC",
                "volume": "",
                "issue": "",
                "pages": "1659--1666",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Joakim Nivre, Marie-Catherine de Marneffe, Filip Ginter, Yoav Goldberg, Jan Haji\u010d, Christopher D. Manning, Ryan McDonald, Slav Petrov, Sampo Pyysalo, Natalia Silveira, Reut Tsarfaty, and Daniel Zeman. 2016. Universal Dependencies v1: a multilingual treebank collection. In Proc. of LREC, pages 1659-1666, Portoro\u017e, Slovenia.",
                "links": null
            },
            "BIBREF9": {
                "ref_id": "b9",
                "title": "All roads lead to UD: Converting Stanford and Penn parses to English Universal Dependencies with multilayer annotations",
                "authors": [
                    {
                        "first": "Siyao",
                        "middle": [],
                        "last": "Peng",
                        "suffix": ""
                    },
                    {
                        "first": "Amir",
                        "middle": [],
                        "last": "Zeldes",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "Proc. of the Joint Workshop on Linguistic Annotation, Multiword Expressions and Constructions",
                "volume": "",
                "issue": "",
                "pages": "167--177",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Siyao Peng and Amir Zeldes. 2018. All roads lead to UD: Converting Stanford and Penn parses to English Universal Dependencies with multilayer annotations. In Proc. of the Joint Workshop on Linguistic Annotation, Multiword Expressions and Constructions (LAW-MWE-CxG-2018), pages 167-177, Santa Fe, New Mexico, USA.",
                "links": null
            },
            "BIBREF10": {
                "ref_id": "b10",
                "title": "Anaphora resolution with the ARRAU corpus",
                "authors": [
                    {
                        "first": "Massimo",
                        "middle": [],
                        "last": "Poesio",
                        "suffix": ""
                    },
                    {
                        "first": "Yulia",
                        "middle": [],
                        "last": "Grishina",
                        "suffix": ""
                    },
                    {
                        "first": "Varada",
                        "middle": [],
                        "last": "Kolhatkar",
                        "suffix": ""
                    },
                    {
                        "first": "Nafise",
                        "middle": [],
                        "last": "Moosavi",
                        "suffix": ""
                    },
                    {
                        "first": "Ina",
                        "middle": [],
                        "last": "Roesiger",
                        "suffix": ""
                    },
                    {
                        "first": "Adam",
                        "middle": [],
                        "last": "Roussel",
                        "suffix": ""
                    },
                    {
                        "first": "Fabian",
                        "middle": [],
                        "last": "Simonjetz",
                        "suffix": ""
                    },
                    {
                        "first": "Alexandra",
                        "middle": [],
                        "last": "Uma",
                        "suffix": ""
                    },
                    {
                        "first": "Olga",
                        "middle": [],
                        "last": "Uryupina",
                        "suffix": ""
                    },
                    {
                        "first": "Juntao",
                        "middle": [],
                        "last": "Yu",
                        "suffix": ""
                    },
                    {
                        "first": "Heike",
                        "middle": [],
                        "last": "Zinsmeister",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "Proc. of the First Workshop on Computational Models of Reference, Anaphora and Coreference (CRAC 2018)",
                "volume": "",
                "issue": "",
                "pages": "11--22",
                "other_ids": {
                    "DOI": [
                        "10.18653/v1/W18-0702"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Massimo Poesio, Yulia Grishina, Varada Kolhatkar, Nafise Moosavi, Ina Roesiger, Adam Roussel, Fabian Simonjetz, Alexandra Uma, Olga Uryupina, Juntao Yu, and Heike Zinsmeister. 2018. Anaphora resolution with the ARRAU corpus. In Proc. of the First Workshop on Computational Models of Reference, Anaphora and Coreference (CRAC 2018), pages 11-22, New Orleans, LA.",
                "links": null
            },
            "BIBREF11": {
                "ref_id": "b11",
                "title": "Udapi: Universal API for Universal Dependencies",
                "authors": [
                    {
                        "first": "Martin",
                        "middle": [],
                        "last": "Popel",
                        "suffix": ""
                    },
                    {
                        "first": "Zden\u011bk",
                        "middle": [],
                        "last": "\u017dabokrtsk\u00fd",
                        "suffix": ""
                    },
                    {
                        "first": "Martin",
                        "middle": [],
                        "last": "Vojtek",
                        "suffix": ""
                    }
                ],
                "year": 2017,
                "venue": "Proc. of the NoDaLiDa 2017 Workshop on Universal Dependencies",
                "volume": "",
                "issue": "",
                "pages": "96--101",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Martin Popel, Zden\u011bk \u017dabokrtsk\u00fd, and Martin Vojtek. 2017. Udapi: Universal API for Universal Dependencies. In Proc. of the NoDaLiDa 2017 Workshop on Universal Dependencies (UDW 2017), pages 96-101, Gothenburg, Sweden.",
                "links": null
            },
            "BIBREF12": {
                "ref_id": "b12",
                "title": "Local and global algorithms for disambiguation to Wikipedia",
                "authors": [
                    {
                        "first": "Lev",
                        "middle": [],
                        "last": "Ratinov",
                        "suffix": ""
                    },
                    {
                        "first": "Dan",
                        "middle": [],
                        "last": "Roth",
                        "suffix": ""
                    },
                    {
                        "first": "Doug",
                        "middle": [],
                        "last": "Downey",
                        "suffix": ""
                    },
                    {
                        "first": "Mike",
                        "middle": [],
                        "last": "Anderson",
                        "suffix": ""
                    }
                ],
                "year": 2011,
                "venue": "Proc. of ACL-HLT",
                "volume": "",
                "issue": "",
                "pages": "1375--1384",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Lev Ratinov, Dan Roth, Doug Downey, and Mike Anderson. 2011. Local and global algorithms for disambiguation to Wikipedia. In Proc. of ACL-HLT, pages 1375-1384, Portland, Oregon, USA. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF13": {
                "ref_id": "b13",
                "title": "FrameNet II: extended theory and practice",
                "authors": [
                    {
                        "first": "Josef",
                        "middle": [],
                        "last": "Ruppenhofer",
                        "suffix": ""
                    },
                    {
                        "first": "Michael",
                        "middle": [],
                        "last": "Ellsworth",
                        "suffix": ""
                    },
                    {
                        "first": "Miriam",
                        "middle": [
                            "R L"
                        ],
                        "last": "Petruck",
                        "suffix": ""
                    },
                    {
                        "first": "Christopher",
                        "middle": [
                            "R"
                        ],
                        "last": "Johnson",
                        "suffix": ""
                    },
                    {
                        "first": "Collin",
                        "middle": [
                            "F"
                        ],
                        "last": "Baker",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Josef Ruppenhofer, Michael Ellsworth, Miriam R. L. Petruck, Christopher R. Johnson, Collin F. Baker, and Jan Scheffczyk. 2016. FrameNet II: extended theory and practice.",
                "links": null
            },
            "BIBREF14": {
                "ref_id": "b14",
                "title": "Mischievous nominal constructions in Universal Dependencies",
                "authors": [
                    {
                        "first": "Nathan",
                        "middle": [],
                        "last": "Schneider",
                        "suffix": ""
                    },
                    {
                        "first": "Amir",
                        "middle": [],
                        "last": "Zeldes",
                        "suffix": ""
                    }
                ],
                "year": 2021,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:2108.12928"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Nathan Schneider and Amir Zeldes. 2021. Mischievous nominal constructions in Universal Dependencies. arXiv:2108.12928 [cs].",
                "links": null
            },
            "BIBREF15": {
                "ref_id": "b15",
                "title": "A gold standard dependency corpus for English",
                "authors": [
                    {
                        "first": "Natalia",
                        "middle": [],
                        "last": "Silveira",
                        "suffix": ""
                    },
                    {
                        "first": "Timothy",
                        "middle": [],
                        "last": "Dozat",
                        "suffix": ""
                    },
                    {
                        "first": "Marie-Catherine De",
                        "middle": [],
                        "last": "Marneffe",
                        "suffix": ""
                    },
                    {
                        "first": "Samuel",
                        "middle": [
                            "R"
                        ],
                        "last": "Bowman",
                        "suffix": ""
                    },
                    {
                        "first": "Miriam",
                        "middle": [],
                        "last": "Connor",
                        "suffix": ""
                    },
                    {
                        "first": "John",
                        "middle": [],
                        "last": "Bauer",
                        "suffix": ""
                    },
                    {
                        "first": "Christopher",
                        "middle": [
                            "D"
                        ],
                        "last": "Manning",
                        "suffix": ""
                    }
                ],
                "year": 2014,
                "venue": "Proc. of LREC",
                "volume": "",
                "issue": "",
                "pages": "2897--2904",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Natalia Silveira, Timothy Dozat, Marie-Catherine De Marneffe, Samuel R. Bowman, Miriam Connor, John Bauer, and Christopher D. Manning. 2014. A gold standard dependency corpus for English. In Proc. of LREC, pages 2897-2904, Reykjav\u00edk, Iceland.",
                "links": null
            },
            "BIBREF16": {
                "ref_id": "b16",
                "title": "Annotating a broad range of anaphoric phenomena, in a variety of genres: the ARRAU Corpus",
                "authors": [
                    {
                        "first": "Olga",
                        "middle": [],
                        "last": "Uryupina",
                        "suffix": ""
                    },
                    {
                        "first": "Ron",
                        "middle": [],
                        "last": "Artstein",
                        "suffix": ""
                    },
                    {
                        "first": "Antonella",
                        "middle": [],
                        "last": "Bristot",
                        "suffix": ""
                    },
                    {
                        "first": "Federica",
                        "middle": [],
                        "last": "Cavicchio",
                        "suffix": ""
                    },
                    {
                        "first": "Francesca",
                        "middle": [],
                        "last": "Delogu",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Kepa",
                        "suffix": ""
                    },
                    {
                        "first": "Massimo",
                        "middle": [],
                        "last": "Rodriguez",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Poesio",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Natural Language Engineering",
                "volume": "26",
                "issue": "",
                "pages": "95--128",
                "other_ids": {
                    "DOI": [
                        "10.1017/S1351324919000056"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Olga Uryupina, Ron Artstein, Antonella Bristot, Federica Cavicchio, Francesca Delogu, Kepa J. Rodriguez, and Massimo Poesio. 2020. Annotating a broad range of anaphoric phenomena, in a variety of genres: the ARRAU Corpus. Natural Language Engineering, 26:95-128.",
                "links": null
            },
            "BIBREF17": {
                "ref_id": "b17",
                "title": "The GUM corpus: Creating multilayer resources in the classroom",
                "authors": [
                    {
                        "first": "Amir",
                        "middle": [],
                        "last": "Zeldes",
                        "suffix": ""
                    }
                ],
                "year": 2017,
                "venue": "Language Resources and Evaluation",
                "volume": "51",
                "issue": "3",
                "pages": "581--612",
                "other_ids": {
                    "DOI": [
                        "10.1007/s10579-016-9343-x"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Amir Zeldes. 2017. The GUM corpus: Creating multilayer resources in the classroom. Language Resources and Evaluation, 51(3):581-612.",
                "links": null
            },
            "BIBREF18": {
                "ref_id": "b18",
                "title": "The Coptic Universal Dependency Treebank",
                "authors": [
                    {
                        "first": "Amir",
                        "middle": [],
                        "last": "Zeldes",
                        "suffix": ""
                    },
                    {
                        "first": "Mitchell",
                        "middle": [],
                        "last": "Abrams",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "Proc. of the Second Workshop on Universal Dependencies (UDW 2018)",
                "volume": "",
                "issue": "",
                "pages": "192--201",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Amir Zeldes and Mitchell Abrams. 2018. The Coptic Universal Dependency Treebank. In Proc. of the Second Workshop on Universal Dependencies (UDW 2018), pages 192-201, Brussels, Belgium.",
                "links": null
            },
            "BIBREF19": {
                "ref_id": "b19",
                "title": "Date and time in Universal Dependencies",
                "authors": [
                    {
                        "first": "Daniel",
                        "middle": [],
                        "last": "Zeman",
                        "suffix": ""
                    }
                ],
                "year": 2021,
                "venue": "Proc. of the Fifth Workshop on Universal Dependencies",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Daniel Zeman. 2021. Date and time in Universal Dependencies. In Proc. of the Fifth Workshop on Universal Dependencies (UDW, SyntaxFest 2021).",
                "links": null
            }
        },
        "ref_entries": {
            "FIGREF0": {
                "uris": null,
                "num": null,
                "text": "Proposed modifier structure for appellationFigure 1",
                "type_str": "figure"
            },
            "FIGREF3": {
                "uris": null,
                "num": null,
                "text": "Phrasal attributive modifiers (hyphen tokens omitted for brevity).",
                "type_str": "figure"
            },
            "TABREF0": {
                "type_str": "table",
                "html": null,
                "num": null,
                "text": "b. French actors Ulliel and Marceau 11 [embellishment] c. Sam and Isaac, my brothers [appos] d. *eggs carton(s)",
                "content": "<table/>"
            },
            "TABREF3": {
                "type_str": "table",
                "html": null,
                "num": null,
                "text": "Constructions involving names and their syntactic properties.",
                "content": "<table/>"
            },
            "TABREF6": {
                "type_str": "table",
                "html": null,
                "num": null,
                "text": "Frequencies of some mischievous nominal constructions in GUM.",
                "content": "<table/>"
            }
        }
    }
}