For Model IM , we varied the number of user intents (K) in intervals from 100 to 400 (see Figure 3), under the assumption that multiple intents would eXist per entity type.

Further modeling the user intent in Model IM results in significantly better performance over all models and across all metrics.

Model IM shows its biggest gains in the first position of its ranking as evidenced by the Prec©1 metric.

Table 2 reports results for Model IM using K = 200 user intents.

3.1 Intent-based Model ( IM )

In this section we describe our main model, IM , illustrated in Figure 1.

Table 1: Model IM : Generative process for entity-bearing queries.

It is very import for Chinese language processing with the aid of an efficient input method engine ( IME ), of which pinyin-to-Chinese (PTC) conversion is the core part.

Meanwhile, though typos are inevitable during user pinyin inputting, existing IMEs paid little attention to such big inconvenience.

In this paper, motivated by a key equivalence of two decoding algorithms, we propose a joint graph model to globally optimize PTC and typo correction for IME .

The daily life of Chinese people heavily depends on Chinese input method engine ( IME ), no matter whether one is composing an Email, writing an article, or sending a text message.

However, every Chinese word inputted into computer or cellphone cannot be typed through one-to-one mapping of key-to-letter inputting directly, but has to go through an IME as there are thousands of Chinese characters for inputting while only 26 letter keys are available in the keyboard.

An IME is an essential software interface that maps Chinese characters into English letter combinations.

We define three main feature categories (1) abstractness and imageability , (2) supersenses, (3) unsupervised vector-space word representations; each category corresponds to a group of features with a common theme and representation.

0 Abstractness and imageability .

Abstractness and imageability were shown to be useful in detection of metaphors (it is easier to invoke mental pictures of concrete and imageable words) (Turney et al., 2011; Broadwell et al., 2013).

Abstractness and imageability .

The MRC psycholinguistic database is a large dictionary listing linguistic and psycholinguistic attributes obtained experimentally (Wilson, 1988).10 It includes, among other data, 4,295 words rated by the degrees of abstractness and 1,156 words rated by the imageability .

(2013), we use a logistic regression classifier to propagate abstractness and imageability scores from MRC ratings to all words for which we have vector space representations.

F: umm, I’m afraid apparant non—sequiturs are always a hazard of doing summaries ;—)

E: I’m just convulsing my thoughts to the irc log

umm, I’m afraid apparant non—sequiturs are always a hazard of doing summaries ;-)

James had not ever had use for something like that so I’m not sure where I would graft that in.

James said that I’m thinking about moving that to on—activation instead of on—startup anyway as it should still work for a main form - but i still wonder if the on—startup parameter issue should be considered a bug — as it shouldn’t choke.

- i’m willing to scrap it if there is a better schema hidden in gnue somewhere :)

For example, Schulte im Walde (2000) uses 153 verbs in 30 classes, and Joanis et al.

However, some of the functions words, prepositions in particular, are known to carry great amount of syntactic information that is related to lexical meanings of verbs (Schulte im Walde, 2003; Brew and Schulte im Walde, 2002; J oanis et al., 2007).

It is therefore unsurprising that much work on verb classification has adopted them as features (Schulte im Walde, 2000; Brew and Schulte im Walde, 2002; Korhonen et al., 2003).

Trying to overcome the problem of data sparsity, Schulte im Walde (2000) explores the additional use of selectional preference features by augmenting each syntactic slot with the concept to which its head noun belongs in an ontology (e.g.

Although the problem of data sparsity is alleviated to certain extent (3), these features do not generally improve classification performance (Schulte im Walde, 2000; J oanis, 2002).

( IM (VB (target))(OBJ))

(VC(VB (target))(OBJ)) (VC(VBG(target))(OBJ)) (OPRD(TO)( IM (VB(target))(OBJ))) (PMOD(VBG(target))(OBJ))

(PRP(TO)( IM (VB (target))(OBJ)))

We compare the results obtained with those obtained by two other systems participating in the KBGen challenge, namely the UDEL system, a symbolic rule based system developed by a group of students at the University of Delaware; and the IMS system, a statistical system using a probabilistic grammar induced from the training data.

System All Covered Coverage # Trees IMS 0.12 0.12 100%

While both the IMS and the UDEL system have full coverage, our BASE system strongly un-dergenerates failing to account for 69.5% of the test data.

In terms of BLEU score, the best version of our system (AUTEXP) outperforms the probabilistic approach of IMS by a large margin (+0.17) and produces results similar to the fully handcrafted UDEL system (-().

Thus, P(Fm,|G) = P(Fm, |Im ,).

P(G) = HWEU P( Im ,).

First, Snow et al.’s model attempts to determine the graph that maximizes the likelihood P and not the posterior P(G Therefore, their model contains an edge prior P( Im ,) that has to be estimated, whereas in our model it cancels out.

arrivaLt ime ).

departure_t ime or ticket price fare .

departure_t ime , and so the node state P : flight .

(O’Donovan et al., 2005; Schulte im Walde, 2006; Erk, 2007; Preiss et al., 2007; Van de Cruys, 2009; Reisinger and Mooney, 2011; Sun and Korhonen, 2011; Lippincott et al., 2012).

Schulte im Walde et al.

K—means and spectral) algorithms (Schulte im Walde, 2006; Joanis et al., 2008; Sun et al., 2008; Li and Brew, 2008; Korhonen et al., 2008; Sun and Korhonen, 2009; Vlachos et al., 2009; Sun and Korhonen, 2011).

Finally, the model of Schulte im Walde et a1.

0 Character 3-gram: Cka+1Ck+2(i — 3 < k< i+m

o Ime ;n(i—4 < m < n < z’+4,0 < nm < 5) matches one entry in the Peking University dictionary:

o (*) Ime ;n(i—4 < m < n < z’+4,0 < n — m < 5) matches one entry in the informal word list:

Previous NLP-related work uses SIF T (Feng and Lapata, 2010; Bruni et al., 2012) or SURF (Roller and Schulte im Walde, 2013) descriptors for identifying points of interest in an image, quantified by 128-dimensional local descriptors.

The USP norms have been used in many previous studies to evaluate semantic representations (Andrews et al., 2009; Feng and Lapata, 2010; Silberer and Lapata, 2012; Roller and Schulte im Walde, 2013).

Such models extract information about the perceptible characteristics of words from data collected in property norming experiments (Roller and Schulte im Walde, 2013; Silberer and Lapata, 2012) or directly from ‘raw’ data sources such as images (Feng and Lapata, 2010; Bruni et al., 2012).

Multi-modal models outperform language-only models on a range of tasks, including modelling conceptual association and predicting com-positionality (Bruni et al., 2012; Silberer and Lapata, 2012; Roller and Schulte im Walde, 2013).

Up to now, such classifications have been used in applications such as word sense disambiguation (Dorr and Jones, 1996; Kohomban and Lee, 2005), machine translation (Prescher et al., 2000; Koehn and Hoang, 2007), document classification (Klavans and Kan, 1998), and in statistical lexical acquisition in general (Rooth et al., 1999; Merlo and Stevenson, 2001; Korhonen, 2002; Schulte im Walde, 2006).

Two large-scale approaches of this kind are Schulte im Walde (2006), who used k-Means on verb subcategorisation frames and verbal arguments to cluster verbs semantically, and J oanis et al.

To the best of our knowledge, Schulte im Walde (2006) is the only hard-clustering approach that previously incorporated selectional preferences as verb features.

SVM-DA: and um Im not sure about the buttons being in the shape of fruit though.

A: and um I’m not sure about the buttons being in the shape of fruit though.

D: Um like I’m just thinking bright colours.

This work was funded by the DFG Research Project Distributional Approaches to Semantic Relatedness (Marion Weller), the DFG Heisenberg Fellowship SCHU-25 80/ 1-1 (Sabine Schulte im Walde), as well as by the Deutsche Forschungsge-meinschaft grant Models of Morphosyntax for Statistical Machine Translation (Alexander Fraser).

(2013); the newspaper data (HGC - Huge German Corpus) was parsed with Schmid (2000), and subcategorization information was extracted as described in Schulte im Walde (2002b).

Briscoe and Carroll (1997) for English; Sarkar and Zeman (2000) for Czech; Schulte im Walde (2002a) for German; Messiant (2008) for French.

Most of these approaches assume that all target verbs are monosemous (Stevenson and Joanis, 2003; Schulte im Walde, 2006; Joanis et al., 2008; Li and Brew, 2008; Sun et al., 2008; Sun and Korhonen, 2009; Vlachos et al., 2009; Parisien and Stevenson, 2010; Parisien and Stevenson, 2011; Falk et al., 2012; Lippincott et al., 2012; Reichart and Korhonen, 2013; Sun et al., 2013).

Moreover, to the best of our knowledge, none of the following approaches attempt to quantitatively evaluate soft clusterings of verb classes induced by polysemy-aware unsupervised approaches (Korhonen et al., 2003; Lapata and Brew, 2004; Li and Brew, 2007; Schulte im Walde et al., 2008).

Schulte im Walde et al.